The Impact of Rapid Structural Analysis in the ConfigurationDevelopment of 2nd and 3rd Generation Launch Vehicles

Abstract

The process by which aerospace vehicles are developed consists of analyses at levels offidelity that increase as the configuration becomes better defined. Early in the development,configuration definition is driven primarily by aerodynamic factors like performance andstability, with highly simplified models of the structural aspects of the vehicle, often in theform of parametric weight equations. It is only after further development, when a singleprimary configuration has been selected, that a finite element model is typically constructedfor structural analysis. Its most obvious output is a more accurate estimate of structuralweight, but another important benefit of this analysis is the identification of structural showstoppers. With conventional configurations such as subsonic aircraft of the body and wingtype, a large body of experience guides designers away from infeasible configurations.However, with innovative and unconventional designs like those for reusable launch vehicles,analytical detection of problems is important.

The barrier to early application of accurate analysis has been the cycle time for the sizing ofa structural model, often as long as six months to a year. Improvements in the structuralanalysis process are dramatically reducing the cycle time for structural sizing, to the pointwhere application of detailed structural analysis is becoming practical early in theconfiguration definition cycle. This paper discusses several technologies that enable thisacceleration of the structural sizing process. They include rapid generation of finite elementmodels (FEMs), parametric modeling, rapid calculation of aerodynamic loads, fasterstructural sizing for strength and for dynamics, rapid estimation of vehicle mass and itsdistribution on the FEM, rapid local sizing of panels, and rapid sizing of the thermalprotection system (TPS).

Introduction

Configurations of aerospace vehicles havehistorically been developed in a sequentialprocess in which the fidelity of analysesincreases as the level of definition of theconfiguration is increased. At the earliest stagesof a program, the gross vehicle characteristics(payload weight, fuel volume, wing area, targetmass fractions, etc.) necessary to meet themission requirements are selected without a realconfiguration definition. Once the range of thegross parameters is defined, a configurationdesigner begins outlining candidateconfigurations within that range. At this stage ofdevelopment, a configuration definition consistsmainly of a three-view sketch of the vehicle, theouter mold line (OML) definition, and a layoutof major systems.

Since a wide variety of possible configurationscan be defined that satisfy the gross vehiclecharacteristics, a set of trade studies or anoptimization is performed to further refine thedesign. At this stage, the vehicle definition is

driven primarily by aerodynamic considerations(aerodynamic performance and stability &control). The aerodynamic analysis used toevaluate configurations is typically a linearaerodynamic code, but in some cases might behigher-order computational fluid dynamics(CFD). Many of the systems requirements (i.e.total fuel volume, engine thrust, actuatorrequirements, etc.) can be estimated based on theaerodynamic calculations. Typically at this stageof vehicle definition, the structural aspects of aconfiguration are estimated using highlysimplified models, such as areal or volumetricweights, or perhaps parametric weight equations.Using processes such as these, a reasonably goodassessment and refinement of the vehicle isusually performed, and a successful vehicleconfiguration is the outcome.

Once the candidate configurations have beennarrowed down to a “best” configuration, a moredetailed assessment of the vehicle is performed.This involves a higher level of fidelity analysisin all disciplines. In the structures area, structuralFEMs are constructed, loads are calculated, andthe structure is sized based on the calculated

loads. Depending on the configuration and themission, other analyses might also be requiredsuch as Thermal Protection System (TPS) sizing,aeroelastic loads, modal frequencies, aeroelasticand aeroservoelastic stability, andaerothermoelastic interactions. There are manyoutputs of this structural analysis. The mostobvious is an estimate of the structural weight ofthe vehicle. This weight estimate is typicallymore accurate than the simplified weightestimates used earlier in configuration definition,and can be used to refine the vehicleperformance and viability assessment.

A much less obvious output of the structuralanalysis and sizing process is the identificationof structural show stoppers. In the developmentof conventional configurations for which a largebody of experience is available (e.g. body andwing subsonic aircraft), it is rare for real showstoppers to appear that make a configurationinfeasible. This is due to the fact thatconfiguration designers have been working withthese configurations for many years, and havethe benefit of having seen many successful (andsome unsuccessful) configurations. For the mostpart, the pitfalls are known. However, whenunconventional configurations are beingdesigned (and certainly all reusable launchvehicles must be considered unconventional atthis time), there is a much smaller heuristicexperience base upon which to draw, and it isvery possible that an unforeseen structural issuewill prove to make the configuration impossible.Since this process has historically taken sixmonths to a year, many man-years of effort couldeasily be wasted on configurations that haveshow stopping structural issues.

It is clear that structural analysis and sizingearlier in the configuration definition cyclewould be valuable, both by increasing thefidelity of the early performance estimates andby identifying show-stopping issues before alarge amount of resources have been committed.In the past, the reason structural analysis andsizing have not been included early in the designcycle is the lengthy structural analysis and sizingcycle time (several components of which aredescribed below). An analysis cycle of sixmonths to a year simply cannot be included inconfiguration trades that may turn around severaltimes a month. Fortunately, improvements instructural analysis processes and approaches aredramatically reducing the cycle time forstructural sizing to the point where it is almost

possible to insert detailed structural sizing intothe early configuration definition cycle.

This paper will discuss several enablingtechnologies for accelerating the structural sizingprocess to the point that it is useful in earlystudies. The key items for making this happencan be classified according to whether theyenhance the sizing process itself or speed up thepreparation of inputs to the sizing process:

Vehicle Sizing Process• Global Sizing• Global/Local Interface• Local Sizing

Rapid Input Generation for Sizing• Parametric Modeling• Rapid FEM Generation• Rapid Loads and Dynamics Analysis• Rapid FEM Mass Estimation and

Distribution

Each of these technologies will be discussed, andthe current state of the art will be assessed. Theongoing development of an integrated rapidstructural analysis and sizing framework willalso be discussed, as well as several applicationsof the sizing process to realistic configurations.

Vehicle Sizing Process

Although the major factors contributing to thecycle time for structural sizing are related to thesetting up of vehicle models and other analysisinputs, it would be helpful to first consider thesizing process itself. The process is showngraphically in Figure 1, and clearly is verymultidisciplinary and involves many steps. Thisreview will provide an understanding of therequirements which the vehicle FEM and theother analysis inputs must satisfy. There are alsofeatures of the sizing cycle where improvementscan reduce cycle time, and these are worthpointing out.

There are, as mentioned previously, variousinstances of structural analysis during thedevelopment of a vehicle. However, thisdiscussion is concerned primarily with theanalysis that is performed when a FEM of thevehicle is ready. The focus of this paper is on thetechnologies that enable this particular form ofanalysis to be applied earlier in the configurationdevelopment cycle, so it is important to clarifywhat it involves.

Global Sizing

The core of the sizing process is the optimizationof the vehicle FEM at a global level, usually inMSC/NASTRAN. Here, the structure is typicallysized to minimize weight while satisfyingconstraints which can include allowable stressand strain in materials, limits on internal runningloads, vibration mode frequencies, flutter speed,and so on. When the global optimizationconverges, a series of analyses follow to checkthings that are not captured in the global FEM.

Since the series of analyses that follow theoptimization of the global FEM are each capableof changing the properties, loads, weight, andgeometry (in the case of TPS offset) of thevehicle model, the global optimization needs tobe repeated. The entire sizing cycle, from globalto local and back, has to be repeated untilconvergence is achieved.

The iterative nature of the sizing process, and theuse of several analysis codes to complete eachiteration, indicate that streamlining the transferof data from one code to the next will make theprocess run faster. In converting the results ofone program to input for another, the less manualintervention necessary the better.

Global/Local Interface

The primary data from the global optimizationthat are used in the local sizing analyses areinternal loads in the sized structure. The loadsare associated with the local structure being sizedin each analysis, and are therefore extracted,averaged, or otherwise processed from globalFEM results. The processed loads areaccompanied by other information about thevehicle structure, such as thickness limits or jointconcepts, that complete the input for the localanalyses.

When the global optimization converges, themodel needs to be updated to obtain the finalloads. Since the structural sizing maysignificantly affect the weight and balance of thevehicle, the first step is to update the distributionof fuel and payload to ensure that rules abouttotal weight and center of gravity (CG) limits aresatisfied. Once this step is complete, theappropriate analyses are performed, using thesized vehicle FEM, to update the external andinternal loads. Depending on the vehicle, thesecan include static analyses, static aeroelastic

analyses, thermal analyses and others. Theseloads are used for various analyses outside theglobal FEM.

On completion of the local sizing, theinformation to be passed back to the global FEMincludes updated properties for finite elements,updated nonstructural mass distributions, andupdated geometry (when TPS sizing isperformed).

Clearly, there is a lot of data transfer between theglobal FEM and various local analysis codes.Much of the translation of FEM results into inputfor these codes depends on FEM characteristics.For example, a code that checks panel stabilityneeds the average loads from the finite elementsthat make up that panel, a code that estimatesjoint weight needs the running load across thejoint, and a code that sizes TPS at a given pointneeds to know the location of the point relativeto the leading edge of the wing and the material,structural concept, and structural sizing results atthat point. Ideally, the transformations to beapplied to FEM results for each code would beset up automatically when the FEM is created.

Local Sizing

The global FEM does not capture the weights ofjoints, fasteners, and other local reinforcementsfor features like hatches and doors, which arecollectively known as non-optimum weights.Using the updated internal loads from theprevious analysis, these features are analyzedand weights are computed. The tool used atBoeing for this process is called XFEMWTS. Theoutput from this step consists of nonstructuralmass increments to be applied to the FEM toaccount for the non-optimum weights.

The global FEM also is not capable of analyzingindividual panels in skins and in substructurewebs for stability, and other detailed checks thatdepend on the structural concept of the panel,such as stringer crippling or shearing in the coreof a sandwich panel. For these checks, Boeingcodes like ADVISOR or commercial codes likeHYPERSIZER are used. These inputs to theseprograms are loads from the FEM that have beenaveraged in some way to correspond toindividual panels (which may consist of dozensof elements in the FEM), and a description of thedesign space in which to operate. The designspace consists of parameters that define thestructural concept of the panel, like stringer

spacing and skin thickness, and the inputspecifies the ranges of values of these parametersthat are to be searched for the lightest design.Different structural concepts can also becompared in this process. The output from theseprograms is the lightest panel that can satisfy allthe checks performed (buckling, crippling, stress,etc.), and the process provides updated propertiesfor the corresponding elements in the FEM.

If rules are available to define what can andcannot be manufactured, the structure can bechecked for them at either the global or locallevel. For manufacturing factors that affect thedetails of the structure, local analyses arenecessary. For vehicle level guidelines, such aslimits on running loads in the vicinity ofmanufacturing splices, constraints areimplemented in the vehicle FEM and addressedduring global optimization.

For hypersonic vehicles and launch vehicles thatemploy TPS, the sizing of the TPS should beintegrated with the structural sizing. Thestructural OML is offset from the aerodynamicOML by the thickness of the TPS, and thisdifference can be significant for structural loadsin certain situations, for example when a wing isthin compared to the TPS. At this step in thesizing cycle, the TPS is sized and its weight ischanged, and the structural OML is updated aswell. In other words, the local of the structure inthe vehicle FEM is changed.

Rapid Input Generation for SizingIt is in the task of setting up the FEM and othersizing process inputs that the most time can beshaved off the sizing cycle. This task is the mostlabor intensive part of the entire structuralanalysis cycle, but that is not the only reason. Ifthe preparation for the sizing process is carriedout well, the process will execute faster. In thedescription of the sizing process it wasmentioned that the execution of the sizing cyclecould be quickened by automating data transferbetween analysis codes in the cycle, and that thetranslation of these data needed information thatcould be provided when the FEM is generated.These facts indicate that technologies that reducethe set up time for the sizing process, and ensurethat the right inputs are available, have greatpromise for reducing the cycle time for structuralsizing to the point where it is practical to apply itearly in the development of a vehicleconfiguration.

Parametric Modeling

If the vehicle sizing process described above isto be applied earlier in the development of aconfiguration, there will be a need for adaptingto changes in the configuration. Even in thetraditional sequence of events, the configurationoften undergoes revisions during and after thestage where an FEM is being used for analysis.Setting up an FEM has traditionally taken solong that it is not possible for structural analyststo keep up with the latest configuration changes.One approach to address this problem is the useof parametric modeling.

The basic idea of parametric modeling is that bydescribing a configuration in terms of basicparameters, and by making an FEM dependenton these parameters, the FEM can then beadapted to a different configuration by changingthe values of the parameters accordingly. This iseasier said than done, because theinterdependencies between parameters can getquite complicated. Parametric modeling has beenimplemented to varying degrees in differentplaces.

Examples of parameters that describeconfigurations include global quantities likewing span, body diameter, and weight, as well asstructural layout descriptors like rib spacing, thenumber of spars, and bulkhead locations. Todefine a configuration in more detail, parameterslike the structural concepts employed in differentparts of the vehicle, the materials used, thelocations of nonstructural masses like actuatorsand landing gear, and the locations of controlsurface hinges may be specified. Some of theseparameters may be defined independently ofothers, or as functions of one or moreparameters. Depending on the vehicle and thelevel of fidelity intended for the FEM, thenumber of parameters and the complexity of therelationships between them can be small or large.

Generally speaking, there seems to be morecapability for parametric modeling in the worldof computer aided design (CAD) than in therealm of finite element preprocessors. There is aparametric modeling facility in UNIGRAPHICS thatallows parameters and relationships betweenthem to be defined by algebraic expressions,which are used to maintain the relationshipswhen individual parameters are changed. Thisfacility is not only capable of generating newCAD models from a set of parameters, but of

updating an existing model when a parameter ischanged as well.

Among graphical finite element preprocessors,IDEAS has a parametric capability that permitsmodels to be updated when parameters arechanged, while PATRAN bases its parametricfeatures on regenerating the model whenever oneor more parameters need to be changed. Thiscapability of PATRAN is based on its so-calledsession files, which store all the commands thatwere executed to create a model in the form offunction calls in PATRAN Command Language,or PCL. When a session file is replayed inPATRAN, the same commands are executed againin the same sequence. If nothing is changed, thesame model is created again, but by changing thearguments passed to the PCL functions, themodel they create can be altered. When theparametric feature of PATRAN is invoked, duringthe creation of the model in the usual interactivemode, the session files that are written havevariables instead of values for quantities thathave been identified as parameters for later use.These session files can be replayed aftersupplying values for the variable parameters tocreate variants of the model.

Even before the implementation of thisparametric feature in PATRAN, session files werebeing used to achieve parametric modelingcapability. With some familiarity with PCL andsession files, this is not a difficult task. This wasdemonstrated by implementing parametricmodeling using session files for an FEM of aBlended Wing Body (BWB) test model knownas the Low Speed Vehicle (LSV), shown inFigure 2. One pleasant surprise in this case wasthat it actually took less time to set up the FEMby focusing on session files that it did by thenormal, interactive way. This was because, whenthe same command needs to be repeated severaltimes, it’s quicker to copy the PCL calls andmodify the arguments in a text editor than it is topoint and click the mouse through all the menuseach time to generate the same session file. Thereal savings in time come when variations of themodel needed to be generated. The amount oftime saved depend on the type of change beingmade. A change in loft shape, for instance,would have required that the model beeffectively re-built from scratch in the traditionalapproach, compared to which replaying thesession files with the new loft was very littlework. At the other extreme, a simple change ofmaterials would be very little work with either

approach, so parametric modeling would savelittle time. When the total effort is considered,including set up time and numerous variations ofthe model, parametric modeling offeredsignificant time savings. The relative effortsrequired for conventional and parametricmodeling for the LSV configuration are shownin Figure 3.

The drawback of parametric modeling is thepossibility of excessive complexity. With adetailed model, when there are many parameters,and they are interrelated, defining therelationships and properly maintaining themrequires organization and a robust, user friendlytool. With the session file approach in PATRAN, itis necessary to be aware of the consequences ofchanging a parameter, not just in the firstcommand that uses it, but far downstream in thesequence of commands being executed. In acomplex model with hundreds of operationsrequired to create it, this task can be daunting. Astools for parametric modeling develop further,the analyst will be able to worry less aboutorganizational details and bookkeeping, andmodels will be easier to make parametric.

Rapid FEM Generation

Whether or not a model is parametric, a majorcontributor to the lengthy cycle time ofconventional structural sizing using FEMs is thepreparation of the FEM. Speeding up the processof building the FEM goes a long way towardsreducing overall cycle time.

An FEM is typically based on geometry data thatdescribe the OML shape, and a definition of thelayout of structural members within it. Initialestimates of structural sizing and of inertiaproperties complete the model. Configurationdesigners are commonly the engineers in chargeof the OML geometry, though it is common forthem to receive input from aerodynamicsengineers at early stages of configurationdefinition. The layout of structure within theOML of the vehicle is determined by theconfiguration designers also. Their definition ofthe configuration is usually in the form of adrawing in a CAD package such as CATIA orUNIGRAPHICS. The initial structural sizing andmass properties of the configuration are providedby weights engineers or the configurationdesigners themselves, but are usually storedoutside the CAD model.

Although CATIA and UNIGRAPHICS have somemeshing capability, the most common practice isto build the FEM in using graphicalpreprocessing software like PATRAN and IDEAS.These programs permit the FEMs to be generatedfor a variety of analysis codes such as NASTRANand ABAQUS. Although there are various toolsavailable to transport the geometry data fromCAD software to the FEM preprocessor, and avariety of intermediate neutral formats in whichthe data can be exported from one program to beimported into another, in practice the importationof CAD geometry into FEM preprocessors isoften problematic. At a minimum, goodcommunication between the CAD engineers andthe FEM builders is required.

To generate an FEM rapidly, it is desirable notonly to avoid a lengthy translation of CADgeometry for an FEM preprocessor, but to also toreduce the time spent preprocessing the FEM.Different approaches to this end are possible.The common feature that the approachesdiscussed here share is that they take advantageof their application to a particular class ofstructures, namely aerospace vehicles, tosimplify the data required to define the structure.Broadly speaking, these vehicles can bedescribed by the OML geometry combined witha substructure layout. Typically, much of thesubstructure is made up of planar members likespars and frames, and therefore the layout of thesubstructure can be defined in a two dimensional(2D) planform. The 2D layout description issimple to define and can be combined with theOML definition to create an FEM, using a tooldeveloped for this purpose.

The above approach has been implemented onboth sides of the CAD to FEM divide. One toolthat operates within UNIGRAPHICS has beendeveloped at Boeing, and it uses a 2D layoutdefinition to break up the CAD geometry intopieces corresponding to the substructureelements. The subdivided geometry can betranslated into PATRAN where it is ready formeshing without a need for time consumingrefinement. This tool is called MANIAC. Anothertool being developed at Boeing worksindependently of the CAD package, andinternally combines the 2D layout definition withOML geometry to produce a three dimensional(3D) FEM. This tool, called RAPIDFEM, willoperate on geometry from various sources, and iscurrently being used with an aerodynamicgeometry definition from a CFD model. The 2D

layout definition is provided to RAPIDFEM in asimple text file, and the FEM it generates isoutput in MSC/NASTRAN format. The modularnature of this tool makes it a simple task toaccommodate the input of different geometryformats, and to generate FEMs for variousanalysis codes. A typical application ofRAPIDFEM to the generation of a global finiteelement model of a wing (including winglet andcontrol surfaces) is shown in Figure 4.

A benefit of using a 2D layout definition togenerate a 3D FEM is the possibility ofintegration with CAD models, such as thosegenerated with the parametric modeling facilityin UNIGRAPHICS. The 3D CAD models inUNIGRAPHICS contain substructure members thatare displayed from data stored internally in 2D,and when the CAD models are variedparametrically, the internal 2D layouts arechanged. These layouts can be exported for usewith tools like RAPIDFEM. The corresponding 3DOML data needed for RAPIDFEM can also beobtained from the 3D CAD model.

A difficulty in using a 2D layout to define aconfiguration is in handling attachments andinterface structures between major componentsof the vehicle. Structure that connects a wing to abody is often unique to each configuration, andalthough it can be an important driver for sizingportions of the wing and the body, it is oftendifficult to create during the process ofprojecting the 2D layout through the 3D OML.The same is true for thrust structures that supportlaunch vehicle engines, which are oftencomprised of trusses. The way such structuresare handled is to generate the rest of the vehiclemodel using the automated process, and to lateradd these structures to it. The addition of theseunique structures to the model can still beautomated, and has been done using PATRANsession files or even FORTRAN codes, but itremains a separate step following the rapid FEMgeneration process. Figure 5 shows an exampleof a full-vehicle finite element model, where thewing is generated by RAPIDFEM, and the bodyand interface structure generated usingtraditional finite element modeling tools.

With a tool like RAPIDFEM, the generation of theFEM from the input data is so fast (literally amatter of a few minutes), that the ability toupdate an existing FEM for a change in aconfiguration parameter is not necessary. Aslong as the OML geometry and the 2D layout

can be updated parametrically, the new FEM isimmediately available.

Rapid FEM generation makes it possible to startthe vehicle sizing cycle sooner. The optimizationmodel for MSC/NASTRAN is provided by the FEMgeneration tool in much less time that ittraditionally takes to set it up. To continue withthe sizing cycle, inputs are needed for the variousother analyses that comprise the completeglobal-local loop. Much of the information thatneeds to be provided to these analytical toolscomes from the FEM. By integrating the rapidFEM generation tool with these other programs,the inputs are set up at the same time that thevehicle FEM is generated. The areas whereintegration of FEM generation with generation ofinput for other pieces of the structural sizingprocess is shown graphically in Figure 6. Thisapproach reduces the set up time for running thevehicle sizing process even more.

Rapid Loads and Dynamics Analysis

Of course, the finite element model and theaccompanying sizing process is of little value ifaccurate sizing conditions are not available in acomparable amount of time. The most importantof these are the critical load conditions. It isfairly simple to rapidly estimate flight loadsbased on low-fidelity aerodynamic theories suchas linear lifting surface or panel codes, or semi-empirical codes such as S/HABP.Unfortunately, the resulting critical loaddistributions are often inaccurate because theflight regimes are often extreme (high anble ofattack, large control surface deflections,transonic mach regime, etc…). Since loads aresuch an integral piece of the sizing process, it isimportant that they be as accurate as possible,and relying on low-fidelity aerodynamic codes isnot desireable.

In the past (and to some extent, currently), therehas been an enormous cost and schedule barrierto using higher fidelity aerodynamics.Experimental aerodynamics is obviously veryslow and costly due to the need to construct windtunnel models, obtain access to a suitablefacility, and reduce the mountains of datagenerated. Computational Fluid Dynamics(CFD) has also been prohibitive due to the largeman-hour requirements to construct theanalytical models (grids), and the largecomputer-hour requirements to generate thesolutions. This problem is magnified by the

need to compute in excess of 1000 aerodynamicsolutions to support a robust critical loadssurvey.

Fortunately, there are several rapidly maturingtechnologies that will dramatically lower thebarriers to rapid high-fidelity loads analysis inthe next few years. These can be broadlygrouped into (1) computing advances, (2)modeling/gridding advances, and (3) overallprocess advances.

Advances in the price/performance ratios ofavailable computing platforms have beenphenomenal, and the trend shows no sign ofreversing. Of particular interest is the use of lowcost “commodity” computers based on thepersonal computer architecture (Intel/AMD) thatis beginning to offer supercomputer performancefor an order of magnitude lower cost than“traditional” high performance computing. Thisis especially exciting in the context of parallelcomputing using PC clusters, which is proving tobe a very effective approach.

In the aerodynamic modeling technology arena,there are two facets that are of interest to theloads process. The first is to define therequirements for a CFD grid for loads analysis,and the second is to generate that grid. A carefulexamination of the quantities of importance tothe loads analyst (and the sensitivities of theseparameters to grid parameters) gives someimportant insights. First, it cannot beoveremphasized that the goals of CFD analysisfor loads calculation are quite different thanthose of CFD analysis for aerodynamicperformance calculation. In loads analysis, theprimary quantities of interest are lifting surfaceand body bending moments, shears, and torques,along with control surface hinge moments. It isoften possible to capture these quantities veryaccurately with a CFD grid much coarser thanthat required for accurately predicting “typcial”aerodynamic performance indicators, such asdrag and maximum lift coefficient. An exampleof this is shown in Figure 7, where a “fine” gridsuitable for aerodynamic performance analysiswas coarsened by a factor of 2 to give a“medium” grid, which was again coarsened togive a “coarse” grid. The medium grid has oneeighth as many grid points as the fine grid, andthe medium grid has 1/64th as many grid points.The body bending moment distributionscomputed using the coarse and medium gridsindicate almost no difference, indicating that the

coarse grid is probably adequate for bodybending loads calculation. While the coarse gridis not adequate for calculating all loadparameters (particularly control surface hingemoments), the medium grid typically gives goodloads results with almost an order of magnitudefewer grid points than required for performancecalculations.

Within the grid generation topic, there are againtwo problems to considered. The first, of course,is the generation of a robust grid around abaseline configuration in a reasonable amount oftime. Using current structured-grid technology,this process can take over a month for a complexconfiguration. However, advances inunstructured-grid technology for viscous flowsolutions will probably provide huge benefitshere, and we can expect to be getting viscousCFD grids of entire configurations in a matter ofdays or hours instead of weeks. The secondgrid-generation aspect that is required for loadsanalysis is grid perturbation. It is often the casein loads analysis that the baseline grid must bemodified slightly to account for control surfacedeflections or aeroelastic deformations. Whilethe modifications are small, they must be donemany times (on the order of thousands), andmust therefore be entirely automated. When thegrid generation technology is sufficiently mature,it may be feasible to simply regenerate the entiregrid for each perturbation, but for the time being,grid perturbation approaches appear to be moreviable. There are several grid perturbationapproaches available [hartwich, farhat, etc…],mostly targeted at aeroelasticity.

The third area of discussion is the overall loadsanalysis process. Current loads analyses areusually based on the concept of an aerodynamic“database” made up of wind tunnel or CFDaerodynamic data. To compute the loads at agiven flight condition, the database isinterrogated for the various pressuredistributions, increments, and corrections thatapply, and these pieces are combined in theappropriate combinations to give an estimate ofthe trimmed, aeroelastic loads for the flightcondition in question. This approach isbeginning to change as a result of thetechnologies described above. It is currentlyfeasible to directly compute aeroelastic, trimmedflight loads as a part of the high-fidelityaerodynamic solution. This process has beenapplied to generate trimmed flight conditions fora re-entry vehicle, in which the flaperon and

ruddervator deflections were dynamicallyupdated as part of the CFD solution to directlygenerate the pressure distribution for a trimmedflight condition at convergence. If thetechnologies discussed above continue toadvance as expected, we may find that directCFD-based trimmed loads analysis can be usedto rapidly and accurately estimate critical flightloads, especially for trajectory-based vehicles.

Rapid FEM Mass Estimation and Distribution

For load conditions involving any acceleration ofthe vehicle, correct inertia properties areimportant. When FEMs are created, there isusually available some sort of estimate fromconfiguration developers of the mass propertiesof the vehicle. If this weights statement is fairlydetailed it can be a significant effort to putnonstructural mass density and concentratednonstructural masses into the FEM to match theestimated weight. Accomplishing this taskmanually involves a lot of calculations usinglocations of FEM nodes, CG locations ofnonstructural masses, areas of FEM elementsover which nonstructural mass is to be smeared,and so on. There are many opportunities forcycle time reduction by automating these tasks.

The estimation of the inertia properties of andFEM is possible with various tools, ranging fromthe analysis code itself (for exampleMSC/NASTRAN), to the finite elementpreprocessor (such as PATRAN), and evenindependent codes (in-house tools have beenimplemented at Boeing). A stand alone programthat can generate a weights breakdown using anFEM as input is a surprisingly useful tool. Inaddition to verifying that added masses had theintended effect on vehicle inertia properties, sucha tool assists in debugging an FEM by indicatingmissing weight, and is useful for tracking theresults of vehicle sizing by revealing the parts ofthe structure that were resized by theoptimization.

Fewer options are currently available todistribute nonstructural mass on an FEM. Thereare in-house programs that process an existingFEM to determine the inertia of subsections ofthe FEM, and then add mass to make them matchtarget values. Such codes take some time to use,since the FEM needs to be broken into groupsdefined by the analyst. A different approach isused in the RAPIDFEM process mentionedpreviously, in which the individual mass items

from the weights statement are input to theprogram, and it places them appropriately in theFEM.

A key point in the RAPIDFEM approach todistributing nonstructural mass in an FEM is theuse of the 2D layout of substructure. This avoidsa common pitfall of concentrated mass elements– if they are placed on structure that isinadequately supported, the FEM computesspurious local vibration modes, excessivedeformations under acceleration, and so on. The2D layout only includes major structuralmembers, all of which are capable of adequatelysupporting the masses of the various itemscontained in the vehicle. By indicating the partsof the 3D FEM corresponding to these pieces ofsturdy substructure, the 2D layout guides theapplication of mass to proper locations in theFEM.

Using the above approach, each mass item in theinput, such as an engine, a control surfaceactuator, or an avionics box, may be describedby several FEM mass elements, but the input isjust the mass of the item and its CG location.Using straightforward logic to match the totalmass and CG of the FEM mass elements to thespecified values completes the task. Much timeis saved by the fact that the input to the processclosely resembles the weights statementprovided by the configuration developers. Anadded advantage is that the masses representing

individual items are traceable, so that items areeasily removed from, or added to, the vehicleFEM.

Conclusions

Reducing structural sizing cycle time permitsFEM based structural sizing to be applied earlyin configuration development, to the benefit ofthe vehicle development effort.

Rapid generation of FEMs and other sizingprocess inputs are key technologies to reduce theset up time as well as execution time for thestructural sizing cycle.

The use of a 2D layout representation of majorstructural components as the reference point forthe structure of a vehicle facilitates rapid FEMgeneration and modification, and can make itpossible for structural analysis cycles to becompleted in the timeframe of a configurationupdate.

ReferencesHartwichFarhatICATSHABPPANAIR

NASTRANSOLUTION 200•Strength•Dynamic Stability•Aeroelastic Loads

Mass, Payload, FuelSpread Sheet Solver

External Loads ILS, Rapid Aero Database,

SOLUTION 144

Internal Loads Thermal LoadsManufacturability

Probabilistic Design

TPS SizingWeight, OML Offset

HYPERSIZER /ADVISOR Local Panel Optimization

XFEMWTSNon-Optimum Weights

Figure 1: Vehicle Sizing Process

FEM Characteristics for BWB Example:• 3 Spars, 7 Ribs, 8 Control Surfaces• 12 Materials, 149 Property Regions• 4500 Grids, 5200 Elements

Figure 2: FEM for Parametric Modeling Illustration

0 50 100 150 200 250 300 350 400

Manhours

Set UpRevised Substructure - Spars RelocatedLoft Shape Changed to Correct Jig ShapeNew Materials Added, Property Regions Revised

Figure 3: Time Saved with Parametric Modeling

Layout

OML

FEM

Figure 4: Rapid FEM Process

Figure 5: A Vehicle FEM, with the Wing from Rapid FEM

NASTRANSOLUTION 200•Strength•Dynamic Stability•Aeroelastic Loads

Mass, Payload, FuelSpread Sheet Solver

External Loads ILS, Rapid Aero Database,

SOLUTION 144

Internal Loads Thermal LoadsManufacturability

Probabilistic Design

TPS SizingWeight, OML Offset

Rapid FEMGenerates Inputfor Various Tools

in One Step

HYPERSIZER /ADVISOR Local Panel Optimization

XFEMWTSNon-Optimum Weights

Rapid FEMFEM and Initial TPS Sizing

Figure 6: Rapid FEM Inputs to Vehicle Sizing Process

Fuselage Vertical Bending vs Fuselage Station

-400,000

-350,000

-300,000

-250,000

-200,000

-150,000

-100,000

-50,000

0

0 50 100 150 200 250 300 350

Fuselage Station [in]

Ben

ing

Mom

ent [

in-lb

s] CoarseMedium

Figure 7: Control Surface Deflection and CFD-Based Trim Process