NASA / T M,2000-210305 Advancement of Bi-Level Integrated System Synthesis (BLISS) Jaroslaw Sobieszczanski- Sobieski Langley Research Center, Hampton, Virginia Mark S. Emiley George Washington University Joint Institute for Advancement of Flight Sciences, Hampton, Virginia Jeremy S. Agte United States Air Force, San Antonio, Texas Robert R. Sandusky, Jr. George Washington University Joint Institute for Advancement of Flight Sciences, Hampton, Virginia December 2000 https://ntrs.nasa.gov/search.jsp?R=20010020389 2018-11-22T17:57:15+00:00Z
Transcript
NASA / T M,2000-210305
Advancement of Bi-Level Integrated
System Synthesis (BLISS)
Jaroslaw Sobieszczanski- Sobieski
Langley Research Center, Hampton, Virginia
Mark S. Emiley
George Washington University
Joint Institute for Advancement of Flight Sciences, Hampton, Virginia
Jeremy S. Agte
United States Air Force, San Antonio, Texas
Robert R. Sandusky, Jr.
George Washington University
Joint Institute for Advancement of Flight Sciences, Hampton, Virginia
Since its founding, NASA has been dedicated tothe advancement of aeronautics and spacescience. The NASA Scientific and Technical
Information (STI) Program Office plays a keypart in helping NASA maintain this importantrole.
The NASA STI Program Office is operated byLangley Research Center, the lead center forNASA's scientific and technical information. The
NASA STI Program Office provides access to theNASA STI Database, the largest collection ofaeronautical and space science STI in the world.The Program Office is also NASA's institutionalmechanism for disseminating the results of itsresearch and development activities. Theseresults are published by NASA in the NASA STIReport Series, which includes the followingreport types:
TECHNICAL PUBLICATION. Reports ofcompleted research or a major significantphase of research that present the results ofNASA programs and include extensivedata or theoretical analysis. Includescompilations of significant scientific andtechnical data and information deemed to
be of continuing reference value. NASAcounterpart of peer-reviewed formalprofessional papers, but having lessstringent limitations on manuscript lengthand extent of graphic presentations.
TECHNICAL MEMORANDUM. Scientific
and technical findings that are preliminaryor of specialized interest, e.g., quick releasereports, working papers, andbibliographies that contain minimalannotation. Does not contain extensive
analysis.
CONTRACTOR REPORT. Scientific and
technical findings by NASA-sponsoredcontractors and grantees.
CONFERENCE PUBLICATION. Collected
papers from scientific and technicalconferences, symposia, seminars, or othermeetings sponsored or co-sponsored byNASA.
SPECIAL PUBLICATION. Scientific,technical, or historical information fromNASA programs, projects, and missions,often concerned with subjects havingsubstantial public interest.
TECHNICAL TRANSLATION. English-language translations of foreign scientificand technical material pertinent to NASA'smission.
Specialized services that complement the STIProgram Office's diverse offerings includecreating custom thesauri, building customizeddatabases, organizing and publishing researchresults ... even providing videos.
For more information about the NASA STI
Program Office, see the following:
• Access the NASA STI Program Home Pageat http'//www.sti.nasa.gov
Robert R. Sandusky, Jr.§George Washington University
Joint Institute for the Advancement of Flight Sciences (JIAFS)[email protected]
Abstract Introduction
Bi-Level Integrated System Synthesis (BLISS) is amethod for optimization of an engineering system, e.g.,an aerospace vehicle. BLISS consists of optimizationsat the subsystem (module) and system levels to dividethe overall large optimization task into sets of smallerones that can be executed concurrently. In the initialversion of BLISS that was introduced and documented
in previous publications, analysis in the modules waskept at the early conceptual design level. This paperreports on the next step in the BLISS development inwhich the fidelity of the aerodynamic drag andstructural stress and displacement analyses wereupgraded while the method's satisfactory convergencerate was retained.
* Manager, Computational AeroSciences, and MultidisciplinaryResearch Coordinator, NASA Langley Research Center, MS 139,Hampton, Virginia, AIAA Fellow
t Graduate Student, NASA Langley Research Center,Multidisciplinary Optimization Branch, MS 139,Member AIAA
1st Lt United States Air Force, San Antonio Air Logistics Center,TX, Member AIAA
§ Professor, George Washington University, Joint Institute forAdvancement of Flight Sciences, Fellow AIAA
BLISS, for Bi-Level Integrated System Synthesis, is anoptimization method for engineering a modular system,e.g., an aerospace vehicle, in which it is beneficial toseparate the design variables and constraints local to amodule from those that govern the entire system. Suchseparation fosters development of a broad workfront of
people and computers, hence it fits well the currenttrends for massively parallel processing in computertechnology and the concurrent engineering style of theworkforce organization.
The focus on dividing the optimization into thesuboptimizations within each module (subsystem, alsocalled the black box) and a coordinating optimization atthe system level places BLISS in the Multidisciplinary
Design Optimization (MDO)toolbox, in the companyof a few other methods that have the same focus as
BLISS but differ in approach. Representative examplesof these methods are the Collaborative Optimization
(CO) (Braun et al, 1965), the Concurrent SubSpaceOptimization (CSSO) (Sobieszczanski-Sobieski, 1988,Bloebaum et al, 1992), and the Concurrent DesignOptimization (Wujek et al, 1995).
The distinguishing features of BLISS are the use of thesystem objective (e.g., the aircraft range) as theoptimization objective in each of the subsystems and atthe system level, and coupling between theoptimizations at the system and subsystem levels bythe optimum sensitivity derivatives with respect to
Thispaperreportson theaboveBLISSupgradeandresultsof thetestingthatadvancethemethodstowardbecomingatoolsuitableforpracticalapplications.Thereportprovidesa synopsisof the BLISSmethod,describesthe salientfeaturesof the two upgradedmodules,presentssatisfactoryconvergenceresults,andsummarizestheBLISSdevelopmentstatusandthefuturedevelopmentdirection.
Notation
ARm - tail aspect ratioARw - wing aspect ratioBB_ - black box
CD- coefficient of dragCf- skin friction coefficient
D - dragESF - engine scale factorh - altitude
k - safety factorL - lift
L/D - lift to drag ratioLHT- horizontal tail location, % mean aerodynamicchord (% MAC)Lw - wing location, % MACM - Mach number
Nz - maximum load factor
R - rangeSFC - specific fuel consumptionSm- horizontal tail surface area
SREr-- wing surface areaT - throttlet/c - thickness to chord ratio
*The 1998a and 1998b references are also available at
http://techreports.larc.nasa.gov/ltrs/
t_- wingbox sandwich face sheets thicknessests,_- wingbox sandwich caliper thicknessesX_ - design variables local to BB_XL, XU - lower and upper bounds on X, side-constraints
W_ - engine weightWF -- fuel weightWT -- total weightY_.j - behavior variables output from BB_ and sent as
A synopsis of BLISS that also appeared in Agte et al,1999 is as follows.
BLISS is a method for optimization of engineering
systems that separates the system-level optimizationfrom potentially numerous autonomous subsystem
optimizations. As shown in Figure 1, it utilizes asystem architecture in which design and behaviorvariables are split into three categories. X-variables arethose design variables optimized at the local level andare unique to each particular subsystem. Behaviorvariables that are output from one subsystem and inputto another are designated Y, and the system-leveldesign variables are specified as Z. System-levelvariables are those shared by at least two subsystems.
Z - system level
(1 _--------[ 12
Subsystem 1 i_g, S _l ubsystem 2
Subsystem 3 _-------@ubsystem 4
I I x4Figure 1: BLISS system structure
After a best guess initialization, the first step in theBLISS begins with the system analysis and sensitivityanalysis in which Y and the derivatives of Y withrespect to Z and X are computed. A linearapproximation to the system objective (an element ofY) as a function of Z and X is established using theabove derivatives. That approximation is adopted asthe objective function in subdomain optimizations that
follownext. In eachsubdomain(module,or blackbox), the Z and Y variablesare frozenand animprovementintheobjectivefunctionis soughtbythelocaloptimizationsthatuselocalX separatelyin eachmodule.ThefrozenZandY areconstantparametersineachmoduleoptimizationandthemoduleoptimizationis followedby computationof thederivativesof theoptimumwithrespectto theseparameters.Thesecondstepachievesimprovementthroughthe system-levelvariablesZ andis linkedto the first stepby thederivativesof optimumwith respectto parametersZandY. Thederivativesareusedto extrapolateeachsubdomainoptimumasa functionof Z andY. ThefunctionalrelationY=Y(Z) is approximatedbyextrapolationbasedonthesystemsensitivityanalysis.Thesestepsalternateuntilconvergence.A flowchartofthemethodisshowninFigure2.
Sle413_ Slea2_
Opportmlty for Con-
cun'ent Processing
initialize X & Z
System Analysisand Updale
[Sensitivity Analysis I Variables
X = X o + AXoI, r
Z =Z0 + AZovr X = X0 + AXovr
Z =Z° + AZ°pT l4 BLISS CYCLE
Figure 2: BLISS Cycle
Note that the output of step 1 is an optimum change inthe local design variables, AXoPT, in the presence ofconstant Z, and the output of step 2 is an optimumchange in system design variables, AZoPT.
In the original version of BLISS the modules showngenerically in Fig. 2 are Propulsion, Aerodynamics,Structures, and Performance whose detailed
input/output variables are identified later. The commondenominator of these modules was the extreme
simplicity of analyses that employed closed-formexpressions for input-to-output mapping. This was sobecause of the need to test the overall procedure
organization and the two-level algorithm convergence atthe initial development stage of a new method withoutbeing encumbered by long turn-around times in themodules. The next logical step in the BLISS
development is to upgrade the fidelity of the moduleswhile holding the overall procedure organizationunchanged.
Upgrades in the BLISS Structures and
Aerodynamics Modules
The modularity of BLISS permits replacing or addingblack boxes to refine or alter the optimization and
analysis tools in each modules allowing the engineerthe flexibility to exercise his judgment. Having toolsof different level of fidelity in the modules enables
applications of BLISS in different design phases. Theadvanced BLISS method incorporates two new modulesthat can be used in lieu of previous black boxes. Thestructures module now can use the Equivalent
Laminated Plate Solution (ELAPS, Giles, 1986) andthe aerodynamics module can use a code calledAWAVE (Harris, 1964) to perform wave drag analysis.
Integration of ELAPS
In the previous application example, BLISS employeda skin-stringer representation of the internal wing box
bays. This model broke the wing down into a threebay wing box whose geometry varied with the taperratio, wing sweep, thickness to chord ratio, wingspan,and aspect ratio, all manipulated as design variables inthe system-level optimization. The displacements,e.g., the wing twist, and stresses, were computed usingsimple, thin-walled box-beam formulas (e.g., Bruhn,1965)
In the BLISS application shown herein, the level ofaccuracy in this module is raised by substituting theprevious model with the Equivalent Laminated PlateSolution (ELAPS) computer code. This code designedwith preliminary design stage calculations in mind iscapable of modeling aircraft wing structures withmultiple trapezoidal segments. The wing structure isrepresented as a plate whose stiffness is set equivalentto that of the original, built-up, structural box of thewing. ELAPS employs a set of displacement fimctionsdefined over each trapezoidal segment and madecompatible in regard to translations and rotations at thesegment junctions. Minimization of the strain energybased on the Ritz method leads to equations fromwhich to calculate static deflections and internal forces.
The latter are then converted to stresses taking intoaccount the details of the wing box built-up cross-section.
The accuracy of the results of ELAPS has been found tobe somewhat below that of finite element codes (Giles,1986) but the ELAPS input is much simpler and fasterto develop. The computation time for an ELAPSmodel is more than an order of magnitude faster thanthat of an equivalent finite element model - an
important feature for a tool to be integrated into anoptimization procedure.
Integratedin BLISS,ELAPSreceivesits inputfromawe-wocessor routine that generates an input file withthe skin thickness, aspect ratio, taper ratio, thickness tochord ratio, sweep, reference area, and aircraft weight.The model used by ELAPS analyzes stress along thesame three bay wingbox configuration used as anexample in the original application of BLISS. Eachwingbox consists of the top and bottom sandwichpanels of different thicknesses and sandwich websidentical in the front and rear of the wingbox. Thefront spar of the wing box is located at 10% of thechord length and the rear spar lies at 70% of the chordlength. Figure 3 depicts the configuration of theELAPS model used by BLISS.
y
Wingbox
/d
at 0.5 chord
Ill X
Figure 3: Wing Model
The top and bottom panels as well as the webs have thethickness of the sandwich face sheets (t) and thesandwich caliper thickness (ts) as design variables, asdepicted in Fig.4. ELAPS models such a built-upstructure by representing each face and the core asseparate elements linked in a common coordinate grid.
[, L ;[O + O
_1--___-- t_2 t_z-'_ ]['_'-
o ¢ Ct j i_ _ o
_ TFigure 4: Wingbox Model
As it was done in the original BLISS implementation,the aerodynamic loads are being generated within thestructures module in the pre-processor to structuralanalysis. To calculate the lift loads on the wing, thewe-processor routine averages spanwise between anelliptical lift distribution and a linear distribution thatreflects the wing chord taper. The elliptical and taperratio based lift distributions for the wing are eachnormalized to contain an area of unity as illustrated inFigure 5. The averaged, spanwise load distribution ismultiplied by the lift required from the wing anddistributed chordwise. The chordwise distribution is a
typical supersonic one with the center of pressurelocated at 50% of the chord. The aerodynamic loaddistribution would be expected to be calculated by anaerodynamics module using a higher fidelity analysis,e.g., a computational fluid dynamics code. Thus, thepresent aerodynamic loads generation is merely aplaceholder for a real aerodynamic loads analysis in afuture BLISS upgrade.
In summary, the structural module employs ELAPS tocalculate the stresses in the wing box for the givenconfiguration, lift distribution, and correspondingconstraints. It also outputs the wing twist and weightand the objective function for the local optimization.The aerodynamics module accepts the output and
models its influence on the aerodynamic response.
Spanwise Location
I • _aper onape . • • ¢,=pt,ca¢ _ Moo,n_ I
Figure 5: ELAPS Lift Distribution
Integration of AWAVE
The cruise segment of the test case mission issupersonic. The original model (Sobieszczanski-Sobieski et al, 1998) used an approximation relying onthe span efficiency factor. That approximation wasreplaced herein with a code, AWAVE, that is astreamlined version of the far-field wave drag program(Harris, 1964). There are two versions of the Harriswave drag program in common use at LaRC. Theoriginal version, described in the reference, treats liftingsurfaces as a series of 3-dimensional solid elements. A
much faster but slightly less accurate version treatslifting surfaces as 2-dimensional panels with finitethicknesses. Due to compensating errors at positiveand negative roll angles of the Mach cutting plane, the
4
panelversiongivesexcellentresultsfor wavedragcoefficients.Theobjectiveofthelast(AWAVE)effortwastodevelopaversionofthewavedragprogramwiththeaccuracyof thesolidelementprogramthatis fasterthan the panel version. The AWAVE codeimplementedcomputesthewavedragon thebasisofthe aircraftcross-sectiondistributionalong thecenterline,henceit requiresdataaboutthe entireconfigurationgeometryto enablethearearuling ofsupersonicbodydesign.
Compared to the original application of BLISS to thesupersonic business jet case, incorporation of ELAPSand AWAVE in BLISS required some changes toconstraints and allocation of the design variables to the
system and subsystem levels.
In the original BLISS, the taper ratio was a localvariable of the structures module. With the integration
of ELAPS and AWAVE, the taper ratio affects both theaerodynamics and structures module. While theaerodynamics module optimization may tend toward ataper ratio to reduce induced drag, the structures
Z - Variables
hREF,S HT ,AR HT, _
z_-----_ Wvo, W o, N z, WBL. CD[V_ N,M<I,]]H
Constants
Range
T -throttle
Am,- tail sweep
Lw-see Figure 1
L m.-see Figure 1
[t]-thickness array,
size lx9
[ts]-thickness array,
size lx9
)_-tap er ratio
D-drag
ESF-eng. scale fact.
L-lift
Nz-max. load fact.
R-range
SFC-spec. fuel cons.
O-wing twist
WE-engine weight
W F-fuel weight
WT-total weight
AR w- wing aspect ratio
ARnT- tail aspect ratio
h-altitude
M- Mach #
SREF-wing surf area
SnT-tail surf area
t/c-thickness/daord
Aw-wing sweep
I I I I I IX Y Z
Figure 6: Data Dependencies for Business JetModel
module may generate a different taper value to reducestresses. To resolve this trade-off, the taper ratio was
raised to a system variable, capable of influencing bothmodules. Figure 6 shows the current black box andvariable interactions.
In this model there are nine system-level Z-variables,each influencing a minimum of two of the subsystems.The local variables of each subsystem are manipulatedonly in the optimization local to that subsystem. Thepropulsion module has the throttle as its sole localvariable. In the present state of BLISS, the range
module is an exception as it performs no optimization.It only evaluates the Breguet range formula. Theaerodynamics module optimizes the local variables ofthe horizontal tail sweep as well as the variables thatplace the wing and tail along the fuselage axis. Thestructural subsystem optimization operates on thesandwich face sheet and caliper thicknesses for the wingcover panels and the webs of the three wingbox bays.
The ten Y-variables noted in the off-diagonal boxes inFigure 6, represent couplings of the black boxes and arecomputed in the system analysis.
BLISSwasoriginallyimplementedin MATLAB5.3.0at both the systemand subsystemlevels. TheMATLAB Optimization Toolbox was used as anoptimizer in the subsystem and system optimizations.In the version of BLISS reported herein, the use ofMATLAB continued as above with the exception of thestructures and aerodynamics modules that incorporatedELAPS and AWAVE, both written in FORTRAN 77.
MATLAB provides a facility to invoke FORTRANfrom a MATLAB code. To exploit that facility, thepreprocessors to both ELAPS and AWAVE werewritten in FORTRAN and converted into MEX-Files
using the MATLAB mex-function (Appendix). Both
AWAVE and ELAPS were then directly called fromwithin the BLISS modules. On the output side,simple post-processing generated outputs in a formatacceptable to the parts of BLISS that remained beingcoded in MATLAB for further analysis. Because of theMATLAB ability to invoke FORTRAN codes, theBLISS upgrading process may continue by addingFORTRAN-coded modules wherever required whileretaining the MATLAB core that executes the methodlogic illustrated by the flowchart in Fig. 1.
Results
BLISS iterations terminate when the change in theaircraft range objective varies less than ten nauticalmiles. This took seven passes through the flowchart inFigure 2. The system-level design variables convergedwithin the first few passes. Further optimizationsfocused primarily on the local variables. Most of thechanges occurred within the structures module wherethe new ELAPS-based optimization kept refining thevariables searching for the best solution. The majorityof the computational time was spent in this module.Table 1 shows the variable progression through theoptimization process.
The table reflects the major trade-offs that occurbetween the wing sweep angle, airfoil thickness ratio,and the wing aspect ratio, all of which govern thestructural weight and drag that, in turn, influence therange. Ultimately, influences of these variables on therange differ in sign, therefore, the procedure seeks acompromise. For example, the wing sweep initiallyincreases to approximately 70 degrees and then falls to40 while the taper ratio decreases to 0.1. The wingreference area rapidly reduces to 200 square feet as thewing aspect ratio is brought down first to 2.5 and thenincreased to 2.607. The wing position is brieflychanged in the fourth cycle but quickly returns to itsinitial value. The wing configuration progression is
var_cycle 1 2 3 4 5 6 7 6
Range(NM) 1051 449; 393E 354_ 339 c 2432 249: 249:
h (ft) 55000 60000 6000C 6000C 6000( 60000 6000C 6000C
M 1.8 2 2 2 2 2 2
ARw 4 2.5 2.607 2.607 2.60; 2.607 2.607 2.607
Aw(° ) 55 40 40.63 40.63 40.6" 40.63 40.63 40.63
SRE F (ftz) 400 200 200 20C 20( 200 200 200
SHT (ft') 150 150 150 150 15( 150 150 150
ARHT 6.5 8.5 8.5 8.5 8._ 8.5 8.! 8.5
'_w 0.2 0.1 0.1 0.1 0._ 0.1 0." 0.1i
Table 1: Supersonic Business Jet Results
depicted in Figure 7. The aircraft finds its optimalcruise conditions after the first cycle of Mach 2.0 at60,000 feet.
2O
18
12
_,oe
_5
Wing Ranform Progression
Cycle 2 Cycles 3_ ycI__
2 4 6 _ 10 12 14 1_ _ 20 22 _4 2_ _a a0 _ _
Distance from Wing Leading Edge (ft)
Figure 7: Wing Planform Progression
The horizontal tail position and geometry stabilize after
the main wing variables reach their settling points. Thetail position varies significantly but settles at a value of100 percent of the mean aerodynamic chord. The tailsweep ends up almost matching the wing sweep but hasa significantly larger aspect ratio. Further analysis ofthe tail may involve incorporating an ELAPS model of
the tail to increasefidelity of analysisin thatcomponent.
Theskin thicknesseschangethroughoutthe processseekingthemaximumof thestructurecontributiontothe rangeunderthestressconstraintsfor the givenconfigurationgeometry,the lattergovernedbytheZ-variables.TheresultinghistogramisseeninFigure8.
4_
35
3
'_25
_2u
1
o5
o
ts3
__ _t3(hner) I
- - t3(rdddle) _
= = 't3(outer)
_ts3 (inner) I
m . ts3 (mddle) ---
- - ,ts3 (outer}
• . -. .........2 3 4 5 6 7 S
BLISS Cycle Number
Figure 8: Plot of Skin Thickness Variation
Though the first cycle was able to converge toreasonable thicknesses, the second through fifth cycleswere unable to satisfy all constraints given the system-level configuration. Then, by the sixth cycle theoptimizer had found a solution that allowed allconstraints to be met and in the seventh cycle it foundthe optimal configuration. Figure 9 shows theprogression of the aircraft Take-Off Gross Weight andits components of empty weight and fuel.
Aircraft Total and Empty Weigh
_Z
2 _ 4 _ 6 7 S
BLISS Cycle Number
Figure 9: Aircraft Weight
Figure 10 depicts a histogram of the aircraft range. Itstarts off at a feasible design point. The cycles twothrough five did not lie within the design space, butBLISS returned to the design space and settled on afeasible design with optimized range.
The last implementation of BLISS to the supersonicbusiness jet test case (Agte, 1999) yielded a range of2,189 nautical miles. With the addition of AWAVEand ELAPS, the more refined analysis increased therange to 2,493 nautical miles.
Distribution of elapsed computing time over the BLISSmodules is displayed in Table 2. It is evident thatmost of the elapsed time is spent in ELAPS but thatwould change drastically if a CFD-level analysis wereused in the aerodynamics module. If BLISS weregrown to the point where all the major modules wouldconsume about equal amount of the elapsed time, thendistributed ,execution on concurrently operatingmachines (or processors within a multiprocessormachine) would radically compress the elapsed time ofthe entire BLISS execution.
I BLISS I ELAPS AWAVE II Percent of Time 8.36% 18978% 1.87%Table 2: Processor Time Use
The next step in the BLISS development is toincorporate additional modules to increase the analysisfidelity. The largest refinement would be expectedfrom adding a computational fluid dynamics code toperform the aerodynamic analysis, including the loads.The propulsion data quality would benefit fromreplacing the current response surface fitted to a look-uptable with a comprehensive engine analysis. Also, thesimple Breguet formula for the aircraft range wouldneed to be replaced by a complete performance analysis
Conclusions and Remarks
Integration of ELAPS and AWAVE into BLISSdemonstrated the modular nature of the method and its
ability to accommodate refinements. Used in a limited
test case of a supersonic business jet design, the two-level optimization in BLISS was effective in satisfyingthe system-level and local constraints while attaining asystem-level objective within a reasonable number of
iterations. Separation of the system-level designvariables from the local ones enabled optimization for asystem-level objective while providing autonomy of thedesign decision and tool choice within disciplinesrepresented in the modules.
The method is open to further upgrades in terms of thefidelity of analysis and optimization techniquesemployed in the modules. In this regard, it is up to theuser to decide on the variety of tools to be integrated inBLISS as needed by the multidisciplinary optimizationtask at hand.
Further advancement of BLISS from its present statusof a method concept demonstrator to a tool useful in
Finally,asthe increasedfidelity of analysesin themoduleswill exactits pricein terms of the computingelapsed time, concurrent execution of the modularanalyses and optimizations will become an attractiveoption.
Acknowledgement
Michael Parks of the George Washington UniversityJIAFS program provided critical help with MATLABrefinements and FORTRAN integration. Hiscontributions were gratefully appreciated.
References
Agte, J.; Sobieszczanski-Sobieski, J.; and Sandusky,R.: "Supersonic Business Jet Design Through Bi-LevelIntegrated System Synthesis," 1999 World AviationConference, SAE 1999-01-5622.
Bloebaum, C. L.; Hajela, P.; and Sobieszczanski-Sobieski, J.: "Non-Hierarchic System decomposition inStructural Optimization; Engineering Optimization,"19, pp 171-186, 1992.
Braun, R. D., and Kroo, I.M.: "Development andApplications of the Collaborative OptimizationArchitecture in a Multidisciplinary DesignEnvironment," SIAM Journal on Optimization,Alexandrov, N, and Hussaini, M. Y., (eds), 1996.
Bruhn, E. F.: "Analysis and Design of Flight VehicleStructures," Tri-State Offset Co., 1965.
Giles, G. L.: "Equivalent Plate Analysis of AircraftWing Box Structures with General PlanformGeometry," NASA TM-87697, March 1986.
Harris, R. V.: "An Analysis and Correlation of AircraftWave Drag," NASA TMX-947, March 1964.
MATLAB Manual, the MathWorks, Inc., July 1993,Version 5.0, 1997, and MATLAB OptimizationToolbox, User's Guide, 1996.
Raymer, D. P.: Aircraft Design: A ConceptualApproach, AIAA Education Series, 1992.
Sobieszczanski-Sobieski, J.: "Optimization by
Decomposition: A Step from Hierarchic to Non-hierarchic Systems," 2nu NASA/AF Symposium on
Recent Advances in Multidisciplinary Analysis andOptimization; Hampton; Virginia, Sept. 1988; NASACP-3031, pp.51-78; also in NASA TM 101494.
Sobieszczanski-Sobieski, J.; Agte, J.; and Sandusky.R.: "Bi-Level Integrated System Synthesis," Proc. 7 tfi
AIAA/USA/NASA/ISSMO Symposium onMultidisciplinary Analysis and Optimization, AIAAPaper No. 98-4916. Also NASA TM-1998-208715,Sept. 1998a; (to appear in AIAA J. Jan. 2000)
Sobieszczanski-Sobieski, J.; Agte, J.; and Sandusky,R.: "Bi-Level Integrated System Synthesis (BLISS)",NASA/TM-1998-208715, August 1998b.
Wujek, B. A.; Renaud, J.E.; Batill, S. M.; andBrockman, J.B.: "Concurrent Subspace Optimizationusing Design Variable Sharing in a Distributed DesignEnvironment," Azarm, S. (ed), Proc. DesignEngineering Technical Conf., Advances in DesignAutomation, ASME DE Vol. 82, pp.181-188, 1995.
Appendix
The Appendix explains the details of the technique usedto integrate a FORTRAN code into MATLAB. Itdescribes the steps required for replacing sections ofBLISS code with new FORTRAN analysis.
MATLAB-FORTRAN Integration
There are a few steps required for integrating aFORTRAN code into the MATLAB environment that
BLISS is currently programmed in. The user mustlocate the place in BLISS where a call is made toanalysis that is to be replaced. Then he must examinethe input and output of the new and old analysis toensure that the remainder of BLISS is capable ofsupplying input the new analysis requires and that thenew analysis produces all the output expected. Then a
pre-processor routine must be created to present thevariable information to the FORTRAN code in an
appropriate format. Finally the data must be harvestedand returned to the MATLAB module from which the
FORTRAN code is being called in a format compatiblewith MATLAB.
8
Pre-processing Data Collection
In pre-processing, BLISS must pass the neededvariables to a routine which will manipulate them intoa form that the FORTRAN code will accept. Theprogrammer must first ensure that the module in whichhe is pre-processing has access to the required variables(i.e., the subsystem must not be using X variablesassigned to other subsystems).
The two FORTRAN codes integrated requiredformatted text input files. The easiest way to preparethese files was to use a FORTRAN subroutine to take
the design variables, configure them in the way neededto represent the geometry or conditions needed by theanalysis, and write the formatted input file for analysis.
In order to pass variables from the MATLABenvironment to the FORTRAN pre-processingsubroutine, MATLAB's mex-function was invoked. In
this function, a standard gateway subroutine is added tothe FORTRAN pre-processing subroutine. This newroutine collects an array of variables passed in fromMATLAB and assigns them to a FORTRAN array.The gateway routine sends these variables into the pre-processing subroutine.
In MATLAB, the user compiles the FORTRAN codeincluding the gateway routine and the pre-processingroutine using the mex command. This creates a mex-file which is treated as a MATLAB function requiringan input array and an output array. The user then placeshis variables for the pre-processing function into anarray and puts this array into the new mex-file. Thissends the variables to the gateway routine whichassigns them to the variables used in the FORTRANpre-processing subroutine. The input file is generatedand control returns to MATLAB.
Program Insertion
Having prepared the data for analysis by the program,the programmer must locate the section of BLISS thathe wishes to upgrade. The previous analysis must beremoved and the code must be placed such that BLISSwill have performed the new analysis and have dataready for later analysis that the user is not replacing.
Having located the desired calling spot and removed thereplaced analysis, the user simply calls the programfrom within BLISS. By previously compiling theFORTRAN code in question, the user calls the programby typing .tprogram_name in the BLISS code whereprogramname is the command that runs the programfrom the operating system. The program then processesthe prepared input file and returns to BLISS.
The final step that the programmer must perform inorder for BLISS to carry on its optimization isharvesting the data produced by the new program.ELAPS and AWAVE both created output files withdata required for BLISS. There are two basic ways tocollect data produced by FORTRAN codes.
The first way is to use a post-processing techniquesimilar to that of pre-processing. The user would createa search algorithm to locate and collect the data fromthe output file. This in turn would be harvested byusing the mex-function to create a gateway between theFORTRAN data collection routine and the BLISS
variables. The programmer would compile the mex-function gateway routine combined with his datacollection routine, run the new mex-file with an arrayprepared to collect the output of the routine, and extracthis data to the array. The programmer would then needto assign the array variables to the proper variables inthe BLISS code. This technique would be best forcases where the user did not have access to the source
code of the FORTRAN program.
In cases where the programmer does have theFORTRAN program's source code and a fair knowledgeof how the program works, he can edit the code tooutput the needed results in a format compatible withMATLAB. If the user can locate the sections of code
that write out the results to the output file, they canchange the code to output to a file with a .m extension.Files in this format are recognized as MATLABprograms that can be called without FORTRANinteraction. By creating .m files with MATLABvariable assignments corresponding to the data that theuser wishes to collect, the programmer simply runsthese MATLAB files after completion of the mainprogram call and the output is already in MATLABformat. This avoids unnecessary data file searching andreduces MATLAB-FORTRAN interaction. After the
data are collected and assigned to the proper variables inBLISS, the procedure would then continue.
Public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing datasources, gathering and maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any otheraspect of this collection of information, including suggestions for reducing this burden, to Washington Headquarters Services, Directorate for Information Operations andReports, 1215 Jefferson Davis Highway, Suite 1204, Arlington, VA 22202-4302, and to the Office of Management and Budget, Paperwork Reduction Project (0704-0188),Washin_lton, DC 20503.1. AGENCY USE ONLY (Leave blank) 2. REPORT DATE 3. REPORT TYPE AND DATES COVERED
December 2000 Technical Memorandum
4. TITLE AND SUBTITLE
Advancement of Bi-Level Integrated System Synthesis (BLISS)
6. AUTHOR(S)
Jaroslaw Sobieszczanski-Sobieski, Mark S. Emiley, Jeremy S. Agte, and
Robert R. Sandusky, Jr.
7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES)
NASA Langley Research Center
Hampton, VA 23681-2199
9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES)
National Aeronautics and Space Administration
Washington, DC 20546-0001
5. FUNDING NUMBERS
WU 509-10-31-03
8. PERFORMING ORGANIZATIONREPORT NUMBER
L-18000
10. SPONSORING/MONITORINGAGENCY REPORTNUMBER
NASA/TM-2000-210305
11. SUPPLEMENTARY NOTES
Sobieski: Langley Research Center; Emiley and Sandusky: George Washington University, JIAFS; Agte: U.S.
Air Force. Paper presented at the 38th AIAA Aerospace Sciences Meeting and Exhibit, January 10-13, 2000,
Reno, NV, Paper No. AIAA 2000-0421.12a. DISTRIBUTION/AVAILABILITY STATEMENT
Unclassified-Unlimited
Subject Category 05 Distribution: Nonstandard
Availability: NASA CASI (301) 621-0390
12b. DISTRIBUTION CODE
13. ABSTRACT (Maximum 200 words)
Bi-Level Integrated System Synthesis (BLISS) is a method for optimization of an engineering system, e.g., an
aerospace vehicle. BLISS consists of optimizations at the subsystem (module) and system levels to divide the
overall large optimization task into sets of smaller ones that can be executed concurrently. In the initial version
of BLISS that was introduced and documented in previous publications, analysis in the modules was kept at the
early conceptual design level. This paper reports on the next step in the BLISS development in which the
fidelity of the aerodynamic drag and structural stress and displacement analyses were upgraded while the
method's satisfactory convergence rate was retained.
14. SUBJECT TERMS
Parallel processing; Optimization by decomposition; Engineering system design
17. SECURITY CLASSIFICATIONOF REPORT
Unclassified
NSN 7540-01-280-5500
18. SECUHITY CLASSIFICATIONOF THIS PAGE
Unclassified
19. SECURITY CLASSIFICATIONOF ABSTRACT
Unclassified
15. NUMBER OF PAGES
1416. PRICE CODE
A0320. LIMITATION
OF ABSTRACT
UL
StanOara Form 298 (Hey. 2-59)Prescribed by ANSI Std. Z-39-18298-102
