MADO - SOFTWARE PACKAGE FOR HIGH ORDER MULTIDISCIPLINARY AIRCRAFT DESIGN AND OPTIMIZATION

Tomasz Goetzendorf-Grabowski , Jacek Mieloszyk , Dawid Mieszalski Warsaw University of Technology, Warsaw, Poland MSP, Drogomy sl, Poland Keywords: optimization, aircraft design, software

Abstract

The vision
1
representation fidelity
verification in flight tests flight testing: - improvement and verification of data, eg. aerodynamic data

Paper presents an overview of MADO, the Multidisciplinary Aircraft Design and Optimization software package. The package is prepared to link existing and modied tools to utilize them next, in an efcient way for optimization. The tools used in the rst stage of development of MADO were presented. Examples of numerical computations are shown - airfoil and the new designed UAVs.
1 Introduction

higher fidelity earlier verification in ground tests

Classical approach
ground-facility testing: - design improvement, - design verification, - data-set generation.

VP

product definition

product development

time

Fig. 1 The idea of cutting time and costs by SimSAC [3] trol issues. This system had to decrease time necessary to obtain wide knowledge about new designed aircraft. MADO goes a step beyond CEASIOM[5]. Collecting experience of SimSAC project authors present proposition to link current tools into global numerical optimization loop. 2 Main idea of MADO Engineers have to deal with increasing amount of data, which forces the designers to focus on repeatable activities, not on creative work they are paid for. Most of these activities can be automated or semi automated depending on the complexity of the task and experience of the designer. Very often the disciplines (e.g. aerodynamics, stability and control, strength analysis, etc.) exist separately. The lack of interfaces between the tools, or only one way connections, cause that translating data from one solver to another con1

Design of new aircraft is always a challenge. Todays market demands highly efcient planes, fullling challenging missions and often contradicting expectations. Thus every project, especially as complex as new design of aircraft, is a set of compromises. Without meeting the goals from many scientic disciplines the aircraft simply wont y. If the design requirements are very demanding it is often hard to propose sufcient project not using numerical optimization techniques [1, 2]. More information on the early stage of the project helps to reduce costs and work time. This is done by improving the process of making condent decisions, which was considered in the SimSAC[3] project (Fig.1). One of the main achievements of SimSAC project was development of CEASIOM[4] package, that links many different tools, from geometry denition, trough aerodynamic computation, mass and inertia analysis, to stability and con-

T.GOETZENDORF-GRABOWSKI, J.MIELOSZYK & D.MIESZALSKI

sumes most of the work time. Thus even simplest optimization is not in use. Thanks to software environment integration it is possible to improve the design process. The key concept lies in connection of main software analyzers (e.g. stability analyzer, aerodynamic solver, etc.) into one structure (Fig.2). The designer can make such connections from one program to other manually, however it still could consume much time. Depending on skills and experience user can add scripts controlling data ow. The solution is developing easy to use, so-called wrappers, that link tools used in design process into globally managed system. Easy way of data transferring relieves the designer and will allow to increase his creativeness. Many starting points are possible to realize the design task, either from CAD geometry to numerical analysis, or in an opposite way. Dening the complicated nonlinear geometry of an aircraft can also be done by authors innovative interface by setting only few design parameters. This method can be treated as fast, high level geometry denition. The package is thought to extend capabilities of existing software by adding additional components - commercial or free, used in the actual institution and linking them into one suite, which can be used for optimization.
OPTIMIZATION TRANSLATORS
AERODYNAMICS - PANUKL - MUNK (inverse problem) - Mass module - Cx0, Cxf... CAD

on CAD geometry if desired. The package contains the component for static and dynamic stability analysis, which also allows to perform ight simulation and estimate performance [6]. In this way we can take into account many project details. All the unique features of presented package, which provide automation, allow to dene optimization task in a relatively easy way. Gradient [7, 8], Monte Carlo [9], Genetic [10] and Swarming [11] optimization algorithms are currently available in the suite. Designer works with user friendly environment, spending less time on repeatable activities, without sacricing exibility and complexity of the task he denes. Over the time spent working with the software user can increase complexity by adding next scientic disciplines and increasing number of design variables. This procedure will lead to better optimized solutions. 2.2 Main components of MADO

VLM

Current version of MADO contains several components, that are the standalone applications. However, they can also run in batch mode, which is very useful to automate optimization process. In the beginning the freeware (or ourself written) applications were taken into account, due to possibility of making necessary changes to run iteration without any prompting. Some commercial applications, which were available for authors were also initially tested with success (VSAERO, VLAERO by AMI Inc.). 2.2.1 XFOIL

CFD

STABILITY & CONTROL PERFORMANCE SDSA

STRENGTH ANALYSIS FEM

Fig. 2 Topology of software components and connections

XFOIL [12] is a well known program for the design and analysis of subsonic isolated airfoils, developed by Mark Drela. Possibility to use XFOIL in batch mode allowed for easy integration of the software in the optimization loop. 2.2.2 PANUKL

2.1 Current development state of MADO The collected software is capable of performing the aerodynamic, structural analyzes basing

PANUKL is a package for aerodynamic analysis of an aircraft using low order panel method. The package was born in mid nineties and was used many times to compute aerodynamic characteristics of an aircraft including stability derivatives
2

MADO - SOFTWARE PACKAGE

Fig. 3 PANUKL package

[13]. The package (Fig.3) is being developed all the time, reaching its maturity. Current version [14] contains preprocessor to dene geometry in an easy way using minimum necessary parameters, vortex wake generator, solver and post-processor to view and analyze the computation results. Although pure potential methods are not able to compute all components of drag, PANUKL can estimate induced drag using Trefftz method. The components of PANUKL package can be run in batch mode, what is useful to automate computations in optimization loop. 2.2.3 SDSA - Simulation and Dynamic Stability Analyzer

SDSA module was developed as the CEASIOM[5] module however it can be run as a standalone application as well. It

was developed for S&C analysis and is able to compute stability characteristics using linear and nonlinear model (simulation model) as well [15]. SDSA uses the same Six DoF mathematical nonlinear model of the aircraft motion for all functions. For the eigenvalue analysis, the model is linearized numerically around the equilibrium (trim) point. Eigenvalues and eigenvectors analysis allow automatic recognition of the typical modes of motion and their parameters. The ight simulation module can be used to perform test ights and record ight parameters in real-time. The recorded data can be used for identication of the typical modes of motions and their parameters (period, damping coefcient, phase shift). The stability analysis results can be assessed basing on CS/FAR, ICAO, and MIL requirements. As a module of MADO, it can receive all the necessary data (aerodynamics, mass, inertia),

T.GOETZENDORF-GRABOWSKI, J.MIELOSZYK & D.MIESZALSKI

Fig. 4 SDSA and its functionality in optimization loop

when available, without special prompting. The necessary data can be delivered to SDSA as an XML le or as a set of plain text les. The second option is useful e.g. for experimental data. The data set contains aerodynamic coefcients or/and stability derivatives tables, mass and inertia data, propulsion data, control derivatives and reference dimensions. The control and propulsion data can be completed and edited using special options of SDSA. SDSA accepts aerodynamic data as tables of stability derivatives as function of angle of attack and Mach number. SDSA also accepts a multidimensional array of force and moment coefcients versus six state parameters (angle of attack, Mach number, sideslip angle and rotational velocity components). A similar array is dened for control derivatives and stability derivatives versus selected accelerations (i.e. alpha dot derivatives). All aerodynamic data (derivatives) can be reviewed and are checked by comparison with typical values. SDSA, running in batch mode, can deliver necessary output data for optimization procedure without any prompting and the optimization process managed by OptoM can run completely in

an automatic way. 2.2.4 OptoM

OptoM (Fig.5) is meant to be easy to use tool for all purpose optimization. Core of the application has incorporated many different optimization algorithms, with possibility of adjusting their settings in graphical user interface. Graphical interface also provides some settings for output format, input and output les paths, ags to use constrains, optimization parameters check, fast update of conguration les, error messages and so on. One of the very useful features is "Tester" option for single objective function analyze, which helps to test and debug optimization task dened by the user. Currently available family of optimization algorithms are: Gradient [7, 8], Monte Carlo [9], Genetic [10] and Swarming [11] methods. All the optimization algorithms can act in signicantly different manner depending on the settings. User of the application has to know only the basics of the optimization algorithms. The main work for him is to properly dene the optimization task in a dynamically linked library,
4

MADO - SOFTWARE PACKAGE

Fig. 5 OptoM screen shot

which is described in more detail in the next chapter. User is also equipped with tools to help him in the denition of the optimization task, like: denition of quadratic penalty function, matrix operations, functions calling external analysis software and many more. 2.2.5 Dynamic linked libraries

Optimization software is part which connects all the MADO tools in one comprehensive aircraft analyze. OptoM has exible structure which is shown on Fig.6. User has majority of optimization algorithms he can choose from, which are incorporated in the main part of the OptoM program. All settings of the optimization algorithms are done in OptoM GUI. Change from one optimization method to another is done by single switch. Providing that the user has previously adjusted settings of the particular optimization algorithm, or agrees for the default settings, optimization process can start immediately with the new algorithm. Optimization task is dened in dynamic library which is linked with the OptoM. Such solution allows designer to focus on de-

nition of his problem to solve without need of deep understanding of how the optimization algorithms work. Only basic knowledge about optimization is needed to adjust the settings if necessary. Ability to dene directly optimization problem in the dynamic library provides exibility and unbounded possibilities for the user. Optimization task can be entirely dened in the dynamic library, but nothing stands against using external analysis software. Any analysis software, which can take input parameters, scripts and commands can be incorporated Fig.6. Only minor difculty is to provide appropriate input for the analysis software and read in results from the analysis during optimization. 3 Numerical examples

This chapter shows two examples of numerical optimization, which utilizes MADO concept. First example concerns of optimization of classical airfoil. The second example concerns MAV optimization.

T.GOETZENDORF-GRABOWSKI, J.MIELOSZYK & D.MIESZALSKI

Fig. 6 OptoM structure

3.1 Airfoil In this example two dimensional airfoil is optimized. Objective function was maximization of lift coefcient for prescribed starting geometry and dened ow conditions. The optimized airfoil was NACA 23012, which had xed angle of attack equal to 3 deg. The airfoil aerodynamic analyze was done by Xfoil code. It was assumed that the airfoil is optimized for small UAV and Reynolds number was set to Re=200000 and Mach number Ma=0. Design variables were: maximum thickness of the airfoil, maximum camber of the airfoil, position of the maximum thickness and position of the maximum camber. Changes of the geometry shape were easily made by using Xfoil tools for airfoil geometry modication. Optimization software always minimizes objective function, to maximize airfoil lift coefcient the objective function was mathematically dened as in equation (1). FOBJ = 1/CL (1)
Fig. 7 Airfoil objective function - maximization of lift coefcient

Fig. 8 History of changing design parameters

For the optimization Monte Carlo optimization algorithm was used. The algorithm was set for ten iterations with fty objective function analysis during every iteration. No additional constrains were dened. Convergence of the objective function (maximized lift coefcient) shows Fig.7. History of changing design parameters is shown on Fig.8. Initial and optimized geometry is shown on Fig.9. After ten iterations airfoil lift coefcient for xed angle of attack improved by dCL = 0.1. In the case of the considered airfoil most inuence on the objective function had parameters of max-

Fig. 9 Airfoil before and after optimization

imum thickness and position of maximum camber and moderate inuence of position of maximum thickness. Denition of the objective function and commands ow to make analysis with Xfoil software took less than one hundred lines of C++ code written in the OptoM shared library. Time of optimization was about fteen minutes.
6

MADO - SOFTWARE PACKAGE

3.2 Bee - mini UAV In the second example minimization of the aerodynamic drag was optimized for constant lift coefcient, with additional constrains put on equilibrium of forces in the aerodynamic z axis and predened static stability margin. Constrains were realized by quadratic penalty function. Achieved values of penalties from the crossed constrains were very big so the objective function was scaled by a factor of one hundred, which is dened by equation (2).

Fob jective = 100CD + P1 + P2 where:

(2)

Fig. 11 Example of competitive optimization solutions center of gravity was added. This way geometry could vary still satisfying the constrains. For optimization genetic and gradient algorithms were used, both giving corresponding results. Genetic algorithm was very robust and always gave solutions. Contrary to that experience gradient algorithm needed quite some time to be set to start the solution to converge, but after it was done it converged much faster than genetic algorithm. The best type of gradient solver was second order Newton method. Although this algorithm needed second order derivatives, its estimation of search direction was so ne, that its efciency overtook Steepest Descent, Conjugate Gradient and Quasi Newton gradient methods. Interestingly genetic algorithm, which can theoretically lead to random solutions, showed that two competitive solutions are possible, with the current optimization task denition, with completely different geometry (Fig.11). The obtained solution, from the rst optimization tests, revealed a question: are the congurations dynamically stable in all modes? The basic tests of dynamic stability were performed and satisfying results were obtained for the rst conguration, with the aft wing sweep (Fig.12). In the next step SDSA is going to run in
7

C1 = mg 0.5 V 2 SCL
2 P2 = 0.5C2 / C2 = 0.1 dCm /dCL

Analysis were done utilizing non-viscous panel code PANUKL[14]. Design variables were: angle of attack, length of the tip chord, three parameters, which controlled nonlinear wing twist distribution dened as a fourth order polynomial center of gravity and four variables dening wing tip cut and let in the middle part of the wing Fig.10.

Fig. 10 Bee wing planform

The rst optimization results showed that the optimization task was over constrained. Solution satised stability constrains by varying angle of attack, but the geometry parameters didnt change signicantly maintaining the old geometry. After this experience variable of position of

T.GOETZENDORF-GRABOWSKI, J.MIELOSZYK & D.MIESZALSKI

zero, which means that both constrains are satised with great accuracy.

Fig. 12 Basic modes of motion for analyzed UAV

the optimization loop to satisfy the stability criteria "on-line". In the nal version authors forecast, that SDSA will be able to modify the decision variables or be a part of constraints system. For the nal optimization, which included more design parameters and geometry details, Swarming optimization algorithm was used. This algorithm was set for ten iterations with fty objective function analysis during every iteration. History of converging objective function is shown on Fig.13. It shows how values of the objective function for the best and the worst individual in the swarm are getting closer. It can be seen, that already in the third iteration objective function, indicating best individual in swarm, merely changes. Performance of the optimization algorithm is very good, what can be observed on the Fig.13 with logarithmic objective function axis.

Fig. 14 Penalty functions values versus iteration number Final geometry with very smooth pressure distribution after optimization shows Fig.15. In the case of the MAV shared library was quite big, but most of the code dened highly nonlinear geometry of the ying object. Pure denition of the objective function and commands ow to execute external aerodynamic analysis took less than one hundred fty lines of code. This example shows that dening optimization task takes less time, but ability to dene the task in the shared dynamic library gives endless exibility and allows for completion of very demanding tasks. 4 Concluded remarks and further steps

Fig. 13 Objective function versus iteration number

On Fig.14 penalty functions values are shown. After the third iteration they are close to

The experience collected so far in the rst examples is very promising. Most of the current MADO tools have user friendly graphical interface with novel functionality, which makes it possible to dene complicated problems for analyze with few variables. Great concern is put on easy data exchange between MADO tools, which allows to concentrate on the very design. Although software maintenance becomes very easy, experienced users will still be able to dene complicated analyze and optimization tasks without limitations. MADO is meant to be so exible, that its working capabilities will grow with the knowledge of the user, no matter on what stage
8

MADO - SOFTWARE PACKAGE

Fig. 15 Pressure distribution for the nal conguration

of the "know how" he is. In the optimization software any desired optimization task can be realized. The optimization task can be dened entirely in the shared library, or incorporate any external analysis software available with batch mode capabilities. Length of the shared library code, with analysis, optimization ow control, objective function, constrains and so on, in most cases will not even exceed 200 lines. Compiling only the shared library, containing only the optimization task and user equipped with additional features to help him dene the task enables for very effective work. Future of the MADO package seems to be bright. The software is constantly developed and improved to make it even more computationally powerful, intuitive and efcient to work with. References
[1] Mieloszyk J., "Handling optimization problems on an example of micro UAV", 3rd CEAS Air&Space Conference, Venice 2011 [2] Stalewski W., Ztak J., "Multi-objective and

[3]

[4]

[5]

[6]

[7]

Multidisciplinary Optimization of Wing for Small Aircraft", 3rd CEAS Air&Space Conference, Venice 2011 SimSAC - "Simulating Aircraft Stability And Control Characteristics for Use in Conceptual Design", Technical Annex to Contract No. FP6030838, July 17, 2006 von Kaenel R., Rizzi A., Oppelstrup J., Goetzendorf-Grabowski T., Ghoreyshi M., Cavagna L., and Berard A. CEASIOM: Simulating Stability & Control with CFD/CSM in Aircraft Conceptual Design, Paper 061, 26th Intl Congress of the Aeronautical Sciences, Anchorage, Alaska, Sept 2008. CEASIOM - Computerised Environment for Aircraft Synthesis and Integrated Optimisation Methods, http://www.ceasiom.com/ SDSA - Simulation and Dynamic Stability analysis, Software package, Warsaw University of Technology, http://www.meil.pw.edu.pl/add/ADD/Teaching/ Software/SDSA Nocedal J., Wright S.J. "Numerical Optimization" Springer, Series in Operations Research, 1999

T.GOETZENDORF-GRABOWSKI, J.MIELOSZYK & D.MIESZALSKI

[8] Vanderplaats G.N.: Numerical Optimization Techniques For Engineering Design, McGraw Hill, 1984 [9] Dickman B.H. and Gilman M.J., Monte Carlo optimization, Journal of Optimization Theory and Applications, Volume 60, Number 1 (1989), pp 149-157 [10] Raymer D.P. "Enhancing Aircraft Conceptual Design Using Multidisciplinary Optimization" Doctoral Thesis, 2002 [11] Li W., Huyse L. and Padula S., Robust airfoil optimization to achieve drag reduction over a range of Mach numbers, Structural and Multidisciplinary Optimization, Volume 24, Number 1 (2002), pp 38-50, Springer-Varlag [12] XFOIL - program for the design and analysis of subsonic isolated airfoils, http://web.mit.edu/drela/Public/web/xfoil/ [13] Goetzendorf-Grabowski T., Goraj Z., Calculations of stability derivatives of an aircraft using panel methods, Transactions of the Institute of Aviation, 2/1996 (145), Warsaw (in Polish) [14] PANUKL potential solver, Software package, Warsaw University of Technology, http://www.meil.pw.edu.pl/add/ADD/Teaching/ Software/PANUKL [15] Goetzendorf-Grabowski T., Mieszalski D., Marcinkiewicz E., Stability analysis using SDSA tool, Progress in Aerospace Sciences, Volume 47, Issue 8, November 2011, Pages 636-646

Copyright Statement
The authors conrm that they, and/or their company or organization, hold copyright on all of the original material included in this paper. The authors also conrm that they have obtained permission, from the copyright holder of any third party material included in this paper, to publish it as part of their paper. The authors conrm that they give permission, or have obtained permission from the copyright holder of this paper, for the publication and distribution of this paper as part of the ICAS2012 proceedings or as individual off-prints from the proceedings.

10