Development of a Multidisciplinary Design Optimization Framework Applied on UAV Design

(1)

Development of a Multidisciplinary

Design Optimization Framework

applied on UAV Design

Athanasios Papageorgiou

Division of Machine Design

Master Thesis

Department of Management and Engineering LIU-IEI-TEK-A-15/02189-SE

(2)

(3)

Development of a Multidisciplinary

Design Optimization Framework

applied on UAV Design

Master Thesis in Multidisciplinary Design Optimization Department of Management and Engineering

Division of Machine Design Linköping University

by

Athanasios Papageorgiou

LIU-IEI-TEK-A-15/02189-SE

Supervisors: Edris Safavi

IEI, Linköping University

Examiner: Kristian Amadori

IEI, Linköping University

(4)

(5)

UPPHOVSRÄTT

Detta dokument hålls tillgängligt på Internet – eller dess framtida ersättare – från publiceringsdatum under förutsättning att inga extraordinära omständigheter uppstår.

Tillgång till dokumentet innebär tillstånd för var och en att läsa, ladda ner, skriva ut enstaka kopior för enskilt bruk och att använda det oförändrat för ickekommersiell forskning och för undervisning. Överföring av upphovsrätten vid en senare tidpunkt kan inte upphäva detta tillstånd. All annan användning av dokumentet kräver upphovsmannens medgivande. För att garantera äktheten, säkerheten och tillgängligheten finns lösningar av teknisk och administrativ art.

Upphovsmannens ideella rätt innefattar rätt att bli nämnd som upphovsman i den omfattning som god sed kräver vid användning av dokumentet på ovan beskrivna sätt samt skydd mot att dokumentet ändras eller presenteras i sådan form eller i sådant sammanhang som är kränkande för upphovsmannens litterära eller konstnärliga anseende eller egenart.

För ytterligare information om Linköping University Electronic Press se förlagets hemsida http://www.ep.liu.se/

C

OPYRIGHT

The publishers will keep this document online on the Internet – or its possible replacement –from the date of publication barring exceptional circumstances.

The online availability of the document implies permanent permission for anyone to read, to download, or to print out single copies for his/hers own use and to use it unchanged for non-commercial research and educational purpose. Subsequent transfers of copyright cannot revoke this permission. All other uses of the document are conditional upon the consent of the copyright owner. The publisher has taken technical and administrative measures to assure authenticity, security and accessibility.

According to intellectual property law the author has the right to be mentioned when his/her work is accessed as described above and to be protected against infringement.

For additional information about the Linköping University Electronic Press and its procedures for publication and for assurance of document integrity, please refer to its www home page: http://www.ep.liu.se/

(6)

(7)

ABSTRACT

The present thesis deals with the topic of Multidisciplinary Design Optimization (MDO) and in particular with the development of a framework for Unmanned Aerial Vehicle (UAV) design. The orientation of the research and the overall outlook of the case-study have been based on Saab’s proposal regarding the application of MDO in complex products and are in compliance with the guidelines of the Innovative Multidisciplinary Product Optimization (IMPOz) project initiative. For this initial stage of the project, five principle engineering disciplines related to aircraft design were considered and those were namely the geometric design, the aerodynamics, the antenna analysis, the Radar Cross Section (RCS) signature, and the mission simulation. The aforementioned disciplines were expressed within the framework by developing computational models which were further based on a relevant set of engineering tools. In the present case-study, the primary focus was on using the indicated engineering tools which are available to both Linköping University and Saab but also to investigate viable and more efficient alternatives. A simple optimization strategy was implemented as a guide for the integration of the models and the core framework configuration was evaluated by using the Design Of Experiments (DOE) method. Finally, the use of metamodels as a tool that can increase the computational efficiency of the framework was analyzed and a preliminary optimization of the product was performed as an example of the framework’s capabilities.

(8)

(9)

ACKNOWLEDGMENTS

To begin with, I would like to thank my examiner Kristian Amadori and the head of the Machine Design division Johan Ölvander for introducing me to the world of MDO and for giving me the opportunity to work on a very interesting project. Your work has been a source of insightful information as well as inspiration and your support was without a doubt instrumental for the completion of this thesis.

Furthermore, a very special “thank you” goes to my supervisor Edris Safavi for all his help throughout the project and especially for his very useful advice and suggestions. I am really grateful for all the time you spent with me and for your guidance in all aspects of the project.

Moreover, I would like to express my appreciation towards the discipline experts from Saab: Carina Marcus, Jakob Bjerkemo, Natalia Gardberg. Thank you for the time you dedicated to this project despite your already busy schedules.

Undoubtedly, a person that deserves my unconditional gratitude is Raghu Chaitanya M.V. of FluMes division. You have been very helpful throughout those six months and your contribution in many technical parts of my project was of vital importance for its completion.

Last but not least, I would like to thank my friends and peers from the M.Sc. program in aeronautics: Aevan, Alejandro and Sharath. You have been very supportive and our discussions as well as exchange of ideas have proven to be far better than any scientific paper or book.

Linköping, June, 2015 Athanasios Papageorgiou

(10)

(11)

ABBREVIATIONS

AC Aerodynamic Center

AKR Anisotropic Kriging

AOA Angle Of Attack

BL Boundary Layer

CAD Computer Aided Design

CFD Computational Fluid Dynamics

CG Center of Gravity

DA Design Automation

DIBA Digital Interactive Basic Aircraftanalysis

DOE Design Of Experiments

DS Direct Sampling (method)

FEM Finite Element Method

GA Genetic Algorithm

GRECO Graphical Electromagnetic Computing

IDS In-Direct Sampling (method)

IMPOz Innovative Multidisciplinary Product Optimization

IR Infrared

IRST Infrared Search and Track

KR Kriging

MAC Mean Aerodynamic Chord

MAV Micro Aerial Vehicle

MAW Missile Approach Warning

MDO Multidisciplinary Design Optimization

ML Multi-level (strategy)

MOGA Multi-Objective Genetic Algorithm

MTOW Maximum Take Off Weight

NP Neutral Point

NRMSE Normalized Root Mean Square Error

RCS Radar Cross Section

RF Radio Frequency

RMSE Root Mean Square Error

SL Single level (strategy)

SM Static Margin

SVD Singular Value Decomposition

UAV Unmanned Aerial Vehicle

UCAV Unmanned Combat Aerial Vehicle

ULH Uniform Latin Hypercube

(12)

(13)

Contributions ... 21 3.1 Framework Evaluation ... 21 3.1.1 Overview ... 21 3.1.2 Problem Formulation ... 22 3.1.3 Design Variables ... 22 3.1.4 Strategy Definition ... 23 3.2 Module Description ... 24 3.2.1 Sizing Loop ... 24 3.2.2 Trim ... 24 3.2.3 Antenna Loop ... 25 3.3 Model Specifications ... 26 3.3.1 Aircraft Geometry ... 26

3.3.2 Aerodynamic Performance By Using VLM ... 28

3.3.3 Aerodynamic Performance By Using CFD ... 30

3.3.4 Antenna Analysis ... 31

3.3.5 Radar Signature ... 32

3.3.6 Mission Simulation ... 33

3.4 Evaluation And Optimization ... 34

3.4.1 Design Of Experiments ... 34

3.4.2 Performance ... 35

3.4.3 Efficient Computing ... 36

3.4.4 Optimization ... 36

Discussion And Conclusions ... 41

4.1 Models ... 41 4.2 Framework ... 43 4.3 Results ... 44 4.4 Summary ... 46 4.5 Future Work ... 47 References ... 49 Appendix A ... 53 Appendix B ... 55 Appendix C ... 57 Appendix D ... 59

(15)

LIST OF FIGURES

Figure 1.1 Overview of similar UAV applications ... 2

Figure 1.2 The multidisciplinary framework of the Saab-IMPOz case-study ... 2

Figure 1.3 Outline of the Thesis ... 6

Figure 2.1 The tail sizing and trim of the aircraft ... 8

Figure 2.2 The geometrical details of the present airfoil definition ... 9

Figure 2.3 The “Demo Gotland” hi-lo-hi mission ... 11

Figure 2.4 The spherical coordinate system ... 14

Figure 2.5 The coordinate transformation for the directive gain ... 15

Figure 2.6 The DS method and the IDS method ... 18

Figure 3.1 The inter-disciplinary relations of the proposed framework ... 23

Figure 3.2 The inter-disciplinary relations and operation of the “sizing loop” ... 24

Figure 3.3 The inter-disciplinary relations and operation of the “trim module” ... 25

Figure 3.4 The inter-disciplinary relations and operation of the “antenna loop” ... 26

Figure 3.5 The aircraft geometry ... 27

Figure 3.6 Geometry parametrization of the engine intake ... 28

Figure 3.7 Example of aperture and sensor placement on the geometry ... 28

Figure 3.8 Result plots from TORNADO ... 29

Figure 3.9 The methodology for importing the airfoil data in TORNADO ... 30

Figure 3.10 The domain shape and size with an overview of the mesh density ... 30

Figure 3.11 An example of the antenna 3D radiation pattern ... 31

Figure 3.12 An example of polar plots of the directive gain ... 32

Figure 3.13 The reference ground radar position and a RCS result plot ... 32

Figure 3.14 The two levels of geometry detail in DIBA ... 33

Figure 3.15 Example of the available mission performance results in DIBA ... 34

Figure 3.16 The optimization of the “sizing loop” with SIMPLEX ... 37

Figure 3.17 Evaluation of MOGA-II for the optimization of the “antenna loop” ... 37

Figure 3.18 The modified framework used in the preliminary optimization ... 38

Figure 3.19 The convergence progress of SIMPLEX when using metamodels ... 38

(16)

(17)

LIST OF TABLES

Table 2.1 Airfoil point coordinates ... 9

Table 3.1 The global and the local optimization problems ... 22

Table 3.2 The global and local design variables ... 23

Table 3.3 The main input and output of the geometry model ... 27

Table 3.4 The main input and output of the TORNADO model ... 29

Table 3.5 The main input and output of the CFD model ... 30

Table 3.6 The main input and output of the antenna model ... 32

Table 3.7 The main input and output of DIBA at a low-fidelity level ... 33

Table 3.8 The main input and output of DIBA at a medium-fidelity level ... 34

Table 3.9 The specifications of the baseline configuration ... 35

Table 3.11 Breakdown of the module performance ... 35

Table 3.12 The obtained NRMSE results from the “antenna loop” metamodel ... 36

Table 3.13 The obtained NRMSE results from the CFD metamodel ... 36

(18)

(19)

I

NTRODUCTION

Part 1 of the thesis introduces briefly the topic of design optimization, the specifications of the IMPOz project, and the details of the present case-study. Furthermore, the applied research methodology is elaborated and the thesis objectives as well as the challenges are presented.

1.1 B

ACKGROUND

The development of complex products is without a doubt a challenging process that involves the interaction of different engineering disciplines. In addition to this, manufacturing industries are oriented towards increased efficiency in all design phases and they strive for time and cost reduction but without losing product quality [1].

Multidisciplinary Design Optimization (MDO) is a promising method which can drastically improve the iterative design process by adding the much needed efficiency in all aspects of the product development process [1], [2]. According to Tarkian [2], “the

implementation of MDO allows the designer to map the interdisciplinary relations that exist in a system and automatically search through the design space for optimal solutions”.

Furthermore, the continuous computational advancements in design tools and the increasing availability of computer power have nowadays made the implementation of MDO feasible even for complex products and during advanced design phases [1], [2].

An important precondition for an effective MDO of a complex system is a Design Automation (DA) framework with parametric capabilities and automated operational features. According to Amadori [1], “DA refers to a system that is able to perform a design

task in an automatic fashion”. More analytically, the framework is able to receive the

design specifications as an input and generate a predefined output which can be subsequently evaluated by the optimizer [1]. As a result, the manual and repetitive tasks which are typically included in the iterative optimization process of a given design are automated and an increased development efficiency as well as a higher design quality can be obtained [2].

1.2 T

HE

IMPO

Z

P

ROJECT

1.2.1 I

N

G

ENERAL

The main objective of the Innovative Multidisciplinary Product Optimization (IMPOz) project is to apply MDO in the manufacturing industry in order to promote more efficient product development processes [3]. The project is under the management of “Kunskapsförmedlingen” (Result Center) which is a Swedish research initiative with focus on product realization, production and support [3]. The research activities emphasize on case-specific products and for this reason there is an ongoing collaboration between Linköping University and industrial partners such as ABB, Bombardier, EnginSoft, Minesto, and Saab.

(20)

According to [3], the basic research topics of the IMPOz project are primarily in respect to the technical and computational requirements of each organization, the partner-specific product design requirements, the available computational tools, and the overall challenges of introducing MDO in the industry. The expected results of the project include guidelines for efficient employment of MDO, optimal product customization, and investigation of possible methods as well as algorithms for MDO [3].

1.2.2 S

AAB

’

S

P

ROPOSAL

The present thesis deals exclusively with the case-study which was proposed by Saab regarding the application of MDO on the development of an Unmanned Aerial Vehicle (UAV). The principle design guidelines include wings of high Aspect Ratio (AR), a V-shaped stabilizer, and symmetry about the XZ (vertical) plane. Similar aircraft which can be considered as reference for the design can be seen in Figure 1.1. Those are the Northrop Grumman RQ-4 Global Hawk, the General Atomics MQ-9 Reaper, and the General Atomics Avenger.

Figure 1.1 Overview of similar UAV applications showing the Global Hawk (left), the Avenger (centre), and the Reaper (right). The photos are courtesy of [4] and [5] respectively.

The research focus of this application is only on certain engineering disciplines, while several other aircraft design aspects are omitted. Figure 1.2 briefly summarizes the multidisciplinary framework which has been considered for the Saab-IMPOz case-study. Some of the models in Figure 1.2 are not included in the scope of the present thesis (marked with blue colour) but they will be included in the framework as part of a future enhancement.

Primary focus within the present case-study is given to topics that are related to the implementation of MDO and in particular to the integration of the tools, to the effectiveness of the applied methodology, to the limitations of this approach, and to the optimization strategy. At the same time, the development of the final product and the complexity of the models are of secondary importance.

(21)

1.3 M

ETHODOLOGY

1.3.1 O

VERVIEW

In order to effectively apply MDO in the development process of the given product, an automated multidisciplinary framework had to be developed. The proposed framework is comprised of several models which correspond to the considered engineering disciplines. Some of the models of the framework are related to only one computational tool but there are also models that require additional support-tools in order to perform the necessary calculations.

The ultimate choice of the computational tool which will represent a model in the framework was based on three basic criteria which are namely the compatibility, the availability, and the fidelity of the software. The term compatibility refers to the ability of the tool to seamlessly communicate with the other components of the framework. By the term availability one refers to whether or not the software can be obtained and used for the current case-study. Finally, the term fidelity is what defines the overall computational capabilities of the software. In a nutshell, low-fidelity tools will give a fast but simplified result, whereas high-fidelity tools are more accurate but come at a higher computational expense.

The models of the proposed framework use tools that are initially suggested by Saab but are also available to Linköping University. In this way, the framework is developed according to the specific technical requirements of the partner-organization. The compatibility between the various tools is not always ideal and therefore, several modifications are required in order to facilitate the interaction between the models. Additional modifications regarding the support of DA features have also been implemented so that the framework can enable a MDO. The overall fidelity of the framework is defined by the preselected tools and it is aimed towards implementing accurate solutions. However, in the case of aircraft aerodynamics, it was decided to test three different options which are different in fidelity and compatibility but are all available to both Linköping University and Saab.

A summary of the research methodology which was implemented in the thesis is presented in the following workflow, while more specific details regarding the development methodology of the individual models and the framework are further elaborated in sections 3.2 and 3.3:

 STEP-1:-Identification of the given and suggested computational tools. Evaluation of aspects such as availability, compatibility and fidelity.

 STEP-2:-Development of the models. Evaluation of different tool alternatives and possible integration solutions. Implementation of modifications so that the models can handle the expected input and deliver the expected output.

 STEP-3:-Definition of the optimization problem, the design variables, and the optimization strategy. Evaluation of the models with respect to handling the selected optimization scheme and implementation of changes in order to effectively handle the inter-disciplinary relations.

 STEP-4:-Integration of the models in the common framework and evaluation of the framework’s performance. Identification of the computationally expensive segments and application of corrective changes.

(22)

 STEP-5:-Further improvement of the framework’s performance by implementing efficient computing solutions. Optimization of the product and evaluation of the optimization scheme and the MDO method.

1.3.2 D

ESCRIPTION

O

F

T

HE

T

OOLS

The basis for the framework of the present case-study was developed by using the commercial software modeFRONTIER by ESTECO [6]. According to the developers, the software is aimed towards MDO and enables the automation of the simulation process as well as the coupling between third-party computer applications.

In the center of the framework is the geometrical model of the UAV which was created in Dassault Systemes CATIA V5R21 design tool suite [7]. The software offers the standard Computer Aided Design (CAD) functions but it is also possible to include knowledge-based features which are essential in a DA framework.

A computational tool that was used in the development of several of the models which are included in the present framework and in many other instances in order to perform support calculations is Mathworks MATLAB R2014b [8]. In particular, the mission simulation tool DIBA, the aerodynamic performance tool TORNADO, and the sensor analysis tool are all MATLAB-based codes. Furthermore, MATLAB is also used for sizing and trimming the aircraft as well as for pre- and post-processing of the input and output of several models.

The model that analyzes the aerodynamic performance of the aircraft has been developed by using two different approaches. As a medium-fidelity approach, the Vortex Lattice Method (VLM) was considered, whereas as a high-fidelity approach two Computational Fluid Dynamic (CFD) solvers were implemented. For the VLM analysis, the MATLAB-based code TORNADO developed by Melin [9] was selected. The CFD codes which are included and tested within the framework are ANSYS Fluent and CFX [10].

The preliminary design toolbox DIBA (Digital Interactive Basic Aircraftanalysis) of version FOI-FFA was used to predict the behavior of the aircraft under specific mission requirements. The tool is provided by Saab and includes several in-house MATLAB codes that can be used to predict the weight, the aerodynamic and the performance characteristics of an aircraft during the conceptual design phase.

For the Radar Cross Section (RCS) signature analysis, the Graphical Electromagnetic Computing (GRECO) method was initially considered. Due to time restrictions and software availability issues the RCS computation was performed by DIBA at a low fidelity level so that the design space can be effectively filled.

1.3.3 C

HALLENGES

A

ND

L

IMITATIONS

A number of challenges were identified throughout the thesis and can be generally placed in two major categories. The first category is about the development challenges of the individual models:

 Although the complexity is of secondary importance, the models should have sufficient fidelity in order to be implemented in a MDO framework.

 The models should be robust and error-free when operating with inputs from the predefined design space of the present case-study.

 The input and output data to and from each model should be manipulated in a compatible form that will allow a smooth flow of information inside the framework.

(23)

 Especially for the geometry model, a proper set of aircraft design rules should be followed and the parametrization should be done in a realistic way that it also corresponds to the specified design variables.

The second category is about the integration of the models in the common framework and the technical requirements that arise from that:

 The framework should be able to clearly depict the chosen optimization strategy but also be flexible to future changes.

 The integration of the models should be robust and at the same time oriented towards increased efficiency.

 The software compatibility issues should be identified and different solutions within the technical limits of the organization should be investigated.

 Computationally expensive segments of the framework must be isolated and possible solutions that can decrease the iteration times should be implemented.

Further challenges include software availability issues and contradictory or unfeasible design demands on behalf of the discipline experts. The latter point is particularly sensitive as it can induce large scale changes at a model- and a framework-level. Good communication, compromise and ingenuity were required throughout the project in order to find an acceptable solution.

1.3.4 R

ESEARCH

Q

UESTIONS

Apart from the manual work that is related to the development of the individual models and their integration in a common framework, the research also focuses on certain topics which are presented below in the form of research questions:

 RQ1:-Are the suggested tools adequate in terms of availability, compatibility, and fidelity?

 RQ2:-How effective are the models with respect to handling the inter-disciplinary relations?

 RQ3:-Is the integration of the models in a common framework under the current optimization strategy feasible?

 RQ4:-Are there any alternative solutions that can increase the efficiency of the framework?

 RQ5:-Is MDO a viable solution for the proposed case-study and what should be the direction of the future work?

1.4 O

BJECTIVES

A

ND

G

OALS

The primary objective of the present thesis is to develop a multidisciplinary framework which will enable a MDO on the design of a UAV. This is further analyzed into two main components which are namely the development of the individual models and the integration of the models into a common DA framework. The models should follow the specifications set in Saab’s proposal and should have adequate complexity in order to support the MDO method. The framework must be above all functional and capable of supporting an optimization strategy which can yield realistic results.

(24)

The primary goal set by the author was to develop the basic models which will comprise the initial framework (see in Figure 1.2) and to integrate them in a common platform. The secondary goal is to test the functionality of the framework through several Design Of Experiments (DOEs), to evaluate the obtained results, and to identify the required iteration times. The third goal is to investigate different methods that can increase the efficiency of the framework and then apply them in order to run a preliminary optimization of the final product.

1.5 R

EPORT

O

UTLINE

The present thesis is organized into four main chapters with the introduction being the first. The outline of the report is graphically summarized in Figure 1.3.

Figure 1.3 Outline of the report

The second chapter includes an overview of the guidelines which were used during the development of the framework models. The scope is to give the reader an overview of the methodology but not to delve into each individual discipline in depth. Throughout the chapter, several relevant case-studies are evaluated and compared to the present application in respect to the most common and practical aspects of MDO.

The third chapter is about the contributions of the present thesis towards the Saab-IMPOz case-study. In this chapter, the proposed framework is elaborated, the methodology and the challenges are described in detail, and the specifications of the individual models are defined.

The fourth chapter is a discussion on the obtained results and an evaluation of the overall project development. Additionally, a summary of the work is offered at the end and the research questions are answered briefly. Finally, the thesis concludes with some suggestions for future work and a description of further research possibilities.

(25)

F

RAME

O

F

R

EFERENCE

Part 2 of the thesis gives a brief overview of some theoretical aspects which are considered relevant to the development of the models and also presents a review of similar design optimization applications.

2.1 A

IRCRAFT

D

ESIGN

2.1.1 W

ING

A

ND

T

AIL

S

IZING

It is a common approach during the definition of the aircraft geometry to use a set of suitable conventions [11], [12], [13], [14]. Formulas (2.1) to (2.3) illustrate how the aspect ratio AR, the taper ratio λ, the span b, the surface area S, the root chord Cr, and the

tip chord Ct are related.

𝐴𝑅 =𝑏2 𝑆 (2.1) 𝜆 =𝐶𝑡 𝐶_𝑟 (2.2) 𝐶_𝑟 = 2 1 + 𝜆√ 𝑆 𝐴𝑅 (2.3)

In the conceptual design phase, empirical formulas are often used in order to size the tail control surfaces [1], [15]. Therefore, the surfaces Sh and Sv of the horizontal and

vertical stabilizer can be approximated by using formulas (2.4) and (2.5) according to both Raymer [11] and Kroo [12]. In order to do that, the volume coefficients Vvt and Vht

must first be defined. In addition to this, the distances lh and lv which are measured from

the Aerodynamic Center (AC) of each stabilizer to the Center of Gravity (CG) of the aircraft (see Figure 2.1) should also be known. However, if the CG is not given, then a good approximation is to consider the distance between the AC of each stabilizer and the AC of the wing [13] [14]. 𝑆ℎ = 𝑉_ℎ𝑡 𝑐̅ 𝑆 𝑙ℎ (2.4) 𝑆𝑣 = 𝑉𝑣𝑡 𝑏 𝑆 𝑙𝑣 (2.5)

For a V-shaped stabilizer the distances lh and lv are clearly equal. Furthermore, the

total surface area S is given by formula (2.6) and the “opening angle” v by formula (2.7).

1

1 𝑆𝑡 = 𝑆ℎ+ 𝑆𝑣 (2.6)

𝑣 = 180 − 2 ∗ tan−1_√𝑆𝑣

𝑆_ℎ

(26)

2.1.2 S

TABILITY

A

ND

T

RIM

The hypothetical longitudinal position of the CG where the aircraft stability is neutral is called the Neutral Point (NP) [13]. The distance from the actual position of the CG xcg to the position of the NP xnp is called the Static Margin (SM) and it is expressed as

a percentage of the Mean Aerodynamic Chord (MAC) c as seen in formula (2.8). 𝑆𝑀 =𝑋𝑛𝑝− 𝑋𝑐𝑔

𝑐 (2.8)

Among other things, the SM defines the pitch stability of the aircraft and hence, it is clear that the design should be within a predefined range [11], [12]. Provided that there is no dynamic model in a MDO framework for predicting the flight characteristics, the SM can be used simply as a constraint like in the case-studies of Choi et al. [16] and Haar et al. [17]. Amadori et al. [18] included the SM in the objective function as a penalty, whereas Lundström et al. [15] approached the stability issue by including an internal loop which balances the aircraft for every optimization evaluation.

Figure 2.1 The basic lengths which are used in the tail sizing and trim of the aircraft, adapted from [13]

An aircraft is considered to be trimmed if there is a balance of forces and moments [11]. For the case of trimming about the pitch axis, a simplified expression for the moments acting about the CG can be seen in formula (2.9). In this formula, Lw and Lht are

the lift forces, while mw and mht are the aerodynamic moments of the wing and the

horizontal stabilizer respectively. The expression neglects thrust terms and assumes that the AC of the wing and tail as well as the CG are all on the same plane [13].

1

1 𝑚𝑤 + 𝑚ℎ𝑡+ 𝐿𝑤𝑥𝑐𝑔+ 𝐿ℎ𝑡𝑙ℎ = 0 (2.9)

During a straight and level cruise the aircraft should be trimmed. For standard aircraft configurations this is achieved by changing the incidence of the horizontal tail stabilizer which in turn affects the drag and hence, the total performance. This issue becomes more complicated if a versatile flight mission is considered where different cruise conditions are expected [19].

For a mission-oriented optimization it is important to calculate the performance based on the correct trim. Lundström et al. [15] addressed this issue in a flying-wing concept by running a panel code aerodynamic analysis at three different Angles Of Attack (AoA) within the linear range of the lift coefficient. In this way, it was possible to get the curves of the lift and the induced drag coefficients as a function of the AoA and therefore, they were able to make performance calculations at various flight conditions. A similar approach but with CFD tools was followed by Hitzel et al. [19] on a standard-tail configuration UAV design and by Haar et al. [17] on a generic aircraft with

(27)

rear-mounted engines. The difference of [19] is that they isolated the horizontal stabilizer from the rest of the aircraft so that they could evaluate it at different AoA. Thus, it was possible to trim the aircraft in the linear region without having to run computationally expensive CFD simulations of the whole aircraft.

2.1.3 A

IRFOIL

D

EFINITION

The airfoil definition and parametrization which is used as reference in the present design is based on the work of Melin et al. [20]. According to the authors, this type of parametric description gives an infinite coordinate resolution that is able to represent an extremely large design space which includes even degenerate profiles. Furthermore, this approach defines the wing profile with only 15 parameters which enables the modeling of many existing airfoils with good tolerance and allows the continuous optimization of the airfoil profile without resulting in discrete curvature changes [20].

Coordinates Points X Y 1 1 A 2 x2 y2 3 x3 B 4 x4 B 5 x5 B 6 0 y6 7 0 0 8 0 y8 9 x9 C 10 x10 C 11 x11 C 12 x12 y12 13 1 A

Table 2.1 Airfoil point coordinates

Figure 2.2 The geometrical details and the control points of the present airfoil definition, adapted by [20]

The airfoil representation is based on four cubic Bezier curves which are defined by 13 control points. The control points are described by their x- and y-coordinates which are subsequently reduced to 14 independent parameters if certain geometrical assumptions are made. The numbering, the order, and the position of the control points can be seen in Figure 2.2, whereas Table 2.1 shows the reduced input parameters. The coordinates for the control points of many known airfoils can be found in the two-part airfoil catalog published by Melin [21], [22].

(28)

2.1.4 R

ADAR

C

ROSS

S

ECTION

The term Radar Cross Section (RCS) refers to the detectability of an object such as an aircraft with respect to one or more defined radar positions [23]. According to Berry [23], the designer should first define the expected mission and thereafter, it is possible to make several geometrical or material choices that can give a lower RCS. Although there is no absolute solution to the problem, some notable stealth geometrical considerations that can be applied in the critical radar directions can be seen below [23]:

 Avoid three- or two-surface reflectors and large vertical flat plates

 Minimize cavities, discontinuities, and straight edges

 Use radar-absorbing materials

The RCS of any target can be described by the term σ which is measured in m2_{and it}

is expressed in spherical directions (see section 2.3.2 for details). A simple definition of the RCS is given in formula (2.10) where r is the examined range, St is the power density

which is intercepted by the target, and Sr is the scattered power density in a distance

equal to r [24]. If the directive gain of an antenna transmitter is known, then a simplification can be made where the RCS can be assumed to be 0.1 m2_{at its peak [25].}

𝜎 = 4𝜋𝑟2𝑆𝑟

𝑆_𝑡 (2.10)

Hitzel et al. [19] did not have a dedicated model for RCS calculations but instead they included the aforementioned design guidelines as geometrical constraints. In particular, they applied an equal minimum sweep to both the wing and tail leading and trailing edges together with some engine placement constraints.

Tianyuan et al. [26] noted the importance of stealth features in military UAV applications and developed a MDO framework that included a model for RCS signature calculations which were used as design constraints. Their RCS analysis tool was based on the “physical optics” theory and it was implemented through a FORTRAN code which required a triangulated surface mesh of the 3D CAD geometry similar to the GRECO method. The authors managed to reduce the computational time by using the same mesh file for both the RCS and the aerodynamic calculations.

A similar need for RCS calculations in military aircraft applications is also expressed in the work of Allison et al. [27]. In this case-study, the authors considered the stealth features early on during the conceptual design as configuration and geometry constraints but they also included a RCS analysis tool in their MDO framework. For the RCS calculations, they used the POFACETS MATLAB-based code [28] together with a meshed 3D geometry. The obtained RCS signature results were used as a comparison between the various designs and were a primary optimization objective.

2.1.5 M

ISSION

P

ERFORMANCE

Two of the most commonly referred performance characteristics are range and endurance. According to Gudmundsson [14], “range is the distance an airplane can fly in

a given time”, whereas “endurance is the length of time an airplane can stay aloft while consuming a specific amount of fuel”. The mathematical expressions for the range R and the endurance E can be seen in formulas (2.11) and (2.12) respectively where V is the

(29)

airspeed, ct is the thrust specific fuel consumption, CL is the lift coefficient, CD is the drag

coefficient, and W is the weight.

𝑅 = ∫ 𝑉 𝑐_𝑡 𝑊𝑖𝑛𝑖 𝑊_𝑓𝑖𝑛 𝐶𝐿 𝐶_𝐷 1 𝑊 𝑑𝑊 (2.11) 𝐸 = ∫ 1 𝑐𝑡 𝑊𝑖𝑛𝑖 𝑊𝑓𝑖𝑛 𝐶_𝐿 𝐶𝐷 1 𝑊 𝑑𝑊 (2.12)

Range and endurance are often encountered in MDO studies as optimization objectives [15], [16], [17], [19], [29], [30] or constraints [18]. The reason for this is that the definitions of both the range and the endurance include aerodynamic and weight parameters but are also integrated for over a mission segment as seen in formulas (2.11) and (2.12). Therefore, it is possible to have a single objective that expresses a design improvement which is related to both structural and aerodynamic performance but also considers the flight conditions [17], [30].

A definition of the term “mission analysis” is given in [14] as “the investigation of an

entire flight of an airplane from engine start to shut-down”. For the purpose of this

analysis, it is a common practice to break the mission into many different parts which will be addressed and analyzed individually [14]. A simple mission consists of the basic segments “taxi and take-off”, “climb”, “cruise”, “loiter”, “descent” and “landing” but there could also be complex missions where the above segments appear more than once (see the example in Figure 2.3).

Figure 2.3 The “Demo Gotland” hi-lo-hi mission which has been considered in the present case-study

The concepts of aircraft performance and mission simulation have been included in many MDO case-studies [15], [16], [18], [19], [27], [31], [30]. The reason is that a given aircraft configuration has different performance characteristics depending on the chosen flight condition [30]. Therefore, in order to get a fully optimized design, it is important to consider all possible flight conditions (mission) rather than isolate just one which will probably lead to a sub-optimal solution or a non-flyable aircraft [27] [31].

Amadori et al. [18], [31] as well as Lundström et al. [15] addressed the issue of mission performance in aircraft MDO by using simple empirical formulas which were included in a dedicated mission simulation module. The aforementioned modules were developed in either MATLAB [31] or MS EXCEL [15] and used the available aerodynamic, propulsion and weight data in order to evaluate the performance for a given mission. In this way, it was possible to check if the concept was capable of flying the mission but also to execute several calculations related to stability and balance.

(30)

The need for feasibility constraints regarding the mission performance is also expressed in the case-study of Allison et al. [27]. For this reason, they included the medium-fidelity mission analysis tool FLOPS [32] which is able to calculate the performance in complex conditions by utilizing the computed outputs from the rest of the framework models. The authors emphasized that an advanced mission analysis tool can add supplementary design fidelity to the framework, however, it should be noted that the same was also reported by [16] and [30] who achieved very good optimization results by using only statistical and empirical methods.

2.2 A

ERODYNAMICS

2.2.1 V

ORTEX

L

ATTICE

M

ETHOD

The Vortex Lattice Method (VLM) is a numerical method which is used in the early aircraft design phases [1]. A typical VLM solver can compute the flow around a complex wing geometry and hence, retrieve the force distribution and the aerodynamic coefficients [33]. The main disadvantages of the VLM is that it neglects the viscous, the compressibility and the body-interaction effects [33].

The TORNADO code was developed by Melin [33] and it is a VLM implemented in

MATLAB. According to the developer [34], the code can solve for most aerodynamic

derivatives and for a wide range of aircraft configurations at a very high computational speed. Over the years, the code has been enhanced with further functions and third-party add-ons which have increased the overall computational fidelity [9]. The features which are most relevant to the present application are the “zero lift drag prediction”, the “consideration of airfoil camber data”, the “Prandtl-Gauert compressibility correction”,

and the “availability of data regarding the International Standard Atmosphere”. A number of MDO frameworks have successfully included TORNADO as the main tool

for aerodynamic predictions mostly due to its fast computational speed and its modifiable interface which enables a simple integration [35], [30]. Smith et al. [30] performed an aero-structural wing optimization, while Safavi et al. [35] used TORNADO for initial aircraft design optimization coupled with a dynamics model. It is noted in [30] that there are limitations at high AoA but nevertheless, the linear aerodynamics theory is valid for normal aircraft operating conditions such as those of cruise, climb, decent, landing, and take-off which have also been considered in the present case-study.

2.2.2 C

OMPUTATIONAL

F

LUID

D

YNAMICS

Computational Fluid Dynamics (CFD) tools implement numerical methods and algorithms in order to resolve a given flow problem [36]. Versteeg et al. [36] state that CFD solvers are high-fidelity computational tools which take into account all aspects of the flow and yield accurate results given that a proper simulation setup has been considered. They are used in later design phases where the design space has been reduced due to their high computational requirements [1]. As Amadori et al. [18] pointed out, the use of a CFD code in very early design phases might be unnecessary as the geometry is still not precisely defined. Furthermore, the authors of [18] note that if the main objective is the comparison of different designs, then CFD can be replaced by a tool of lower fidelity in order to speed up the optimization process.

(31)

CFD tools have been implemented in a variety of MDO applications and it is commonly accepted that they are an accurate but computationally expensive option [16], [17], [18], [19]. Choi et al. [16] used ANSYS Fluent to create a metamodel which they coupled with other low- or medium-fidelity tools. They validated the obtained optimization results with low- and high-fidelity methods and resulted that their approach can increase the accuracy of the framework at an acceptable design cost. Haar et al. [17] as well as Hitzel et al. [19] used the CFD solver TAU [37] for optimization of aircraft designs directly and without the use of metamodels. They comment mostly on the challenges of the integrating a LINUX-based software and on the setbacks that arise from the high failure rate (average of 25 %) which they observed due to CFD process errors. Finally, a more detailed description of the challenges as well as the possible integration methodologies regarding MDO and CFD that can increase the overall efficiency of the framework can be found in the work of Breitkopf [38] but will not be repeated here. Among the many topics that are analyzed, the author points out the advantages of metamodeling as an alternative to CFD and gives examples of successful aircraft design optimizations where the results are “quite close” to the real simulations.

2.3 A

NTENNAS

A

ND

S

ENSORS

2.3.1 O

VERVIEW

In general, an antenna or a sensor is a unit which is comprised of two components, namely the aperture and the electronics [25]. For an antenna that operates in the Radio Frequency (RF), it can be assumed that the electronics and the aperture are separate components. For Infrared (IR) sensors, the electronics and the aperture are usually integrated into one part but there could be designs where the aperture signal is collected in a central processing unit.

In UAV applications the communications are mostly achieved through RF antennas in different frequency bands, whereas the IR sensors are used to increase the situational awareness. For the coverage of the RF bands, it is common to use “aperture antenna” type which can be easily mounted on the aircraft surfaces [24]. The considered IR sensors for the present case are the Infrared Search and Track (IRST) and the Missile Approach Warning (MAW) systems. A small contribution to the overall radar signature of the aircraft is anticipated from the RF antennas but not from the IR sensors which operate as passive systems [25].

2.3.2 B

ASIC

P

RINCIPLES

An important parameter that defines the field performance of an antenna or a sensor is the radiation pattern. According to Balanis [24], “the radiation pattern is defined as a

mathematical representation of the radiation properties as a function of the space coordinates”. Another fundamental parameter is the directivity which indicates the

antenna or sensor capabilities over a certain direction [39]. Balanis [24] defines directivity as “the ratio of the radiation intensity in a given direction to the radiation

intensity averaged over all directions”. A proper set of coordinates is necessary for the

representation of the antenna and sensor properties. The spherical coordinate system (see Figure 2.4) is the most popular, since the interest is usually on a certain direction and at a certain distance from the source [39].

(32)

Figure 2.4 The spherical coordinate system which is used in the analysis of antennas, adapted from [24]

A simple methodology which is used for the modeling of rectangular apertures is elaborated below. The variables that characterize an aperture are the frequency v as well as the dimensions a and b. First, the wavenumber k is calculated based on formulas (2.13) and (2.14).

𝑣 =1

𝜆 (2.13)

𝑘 =2𝜋

𝜆 (2.14)

Secondly, the far zone fields X and Y are defined as a functions of the azimuth and elevation angles φ and θ as shown in formulas (2.15) and (2.16).

𝑋 =𝑘𝑎

2 sin 𝜃 cos 𝜑 (2.15)

𝑌 =𝑘𝑏

2 sin 𝜃 cos 𝜑 (2.16)

Thirdly, the far-zone electric fields Eθ and Eφ are calculated by formulas (2.17) and

(2.18) where C is the maximum amplitude.

𝐸_𝜃 = 𝐶 sin 𝜑sin 𝑋 𝑋 sin 𝑌 𝑌 (2.17) 𝐸𝜑 = 𝐶 cos 𝜃 cos 𝜑 sin 𝑋 𝑋 sin 𝑌 𝑌 (2.18)

The radiation intensity U is then given by formula (2.19) where η is the intrinsic impedance of the medium.

𝑈(𝜃, 𝜑) = 1

2𝜂[(𝐸𝜃)2+(𝐸𝜑)

2

(33)

Finally, the directive gain Dg can be determined by formula (2.20) where Prad is the

total radiated power in all directions. 𝐷𝑔 =

4𝜋𝑈

𝑃𝑟𝑎𝑑 (2.20)

The characteristics of the IR sensors are computed in a similar way as shown above with the only difference being the definition of the far-zone electric fields. Those can be modelled according to formulas (2.21) and (2.22).

1

1 𝐸𝜃 = 𝐶 cos 𝜃 (2.21)

1

1 𝐸𝜑= 𝐶 sin 𝜃 (2.22)

2.3.3 C

OORDINATE

T

RANSFORMATIONS

The formulas of section 2.3.2 calculate the radiation pattern of each aperture on a set of local spherical coordinates. In order to enable a holistic quantification of the coverage, the directive gain of each aperture must be expressed on a global spherical coordinate system. A simple graph of the transformation methodology which is also applied in the present model can be seen in Figure 2.5.

Figure 2.5 The coordinate transformation for the directive gain of a typical rectangular aperture

2.3.4 C

ONSIDERATIONS

F

OR

MDO

The coverage and the radar signature of antennas and sensors are important characteristics which can drive the design of UAVs. During the conceptual design phase it is considered acceptable to make several simplifications and assumptions in order to enable a MDO [25]. Thus, the terms “coverage” and the “radar signature” can be substituted by the directive gain, normalized to its maximum amplitude, and multiplied with a suitable amplitude factor.

As far as the power is concerned, one may assume that a sensor requires 1000W of power of which half is used for cooling and the other half is transmitted through the aperture [25]. The aperture is assumed to be a flat-shaped patch that adds little weight (0.3 kg/aperture), while each sensor can be considered to be a box with 1.5x2x3 dm sides and with a mass density of 3 kg/dm3 _[25].

(34)

2.4 M

ULTIDISCIPLINARY

F

RAMEWORKS

2.4.1 O

VERVIEW

The development of an aircraft is a complex process that involves many engineering disciplines with complicated inter-dependencies which means that a holistic MDO can be a quite challenging task. One common element of all the case-studies that are referenced in this thesis is the integration of a CAD geometry in the MDO framework. The other trend that can be observed is that an aerodynamic analysis is included in the majority of the cases. Finally, it can be seen that depending on the application, several disciplines are often omitted or are expressed by very simple (empirical) formulas.

The importance of the CAD geometry in MDO frameworks regarding aircraft design is analyzed by Amadori [1] who notes that the “geometry in-the-loop” approach can be very computationally expensive but at the same time it is extremely powerful with respect to modelling and analysis. Moreover, a CAD geometry can also facilitate the integration of other high-fidelity tools inside the framework like in the case of FEM and CFD.

Aerodynamics are also instrumental in aircraft studies and this is mainly because the obtained coefficients are used to define the structural loads, the trim requirements, the propulsion, and the mission performance [27]. Therefore, it is clear that an aerodynamics model can affect the overall quality of the design. Hence, it should not come as a surprise that there is a tendency to use higher fidelity tools and that authors strive for higher quality results [19].

The rest of the disciplines which will comprise a MDO framework depend mostly on the mission of the aircraft, the targeted design space, and the available computational resources [27]. In many instances, handbook formulas can be used in order to create a simple model just for the purpose of closing the design space and to provide support to the other models [18]. Another characteristic that is emphasized by many authors is the flexibility of the framework which can enable the seamless continuation of the research [15], [16], [17], [18], [26], [27], [29], [31]. In simple terms, a flexible framework should be able to adapt to a future refinement of the models by tools of different fidelity and allow the addition of new models.

2.4.2 O

PTIMIZATION

S

TRATEGIES

One of the possible categorizations that can be found in some of relevant literature samples regarding the optimization methods is the Single Level (SL) and Multi Level (ML) strategies which are mainly based on the number of the involved optimization processes [1], [2]. A SL strategy has a single optimizer, while in a ML strategy the optimization process is distributed in many segments of the framework which according to Tarkian [2] can be a more suitable approach for complex engineering products.

A ML optimization strategy is elaborated in the study of Tianyuan [26] regarding the optimization of UCAV. Initially, an optimization is performed on the aerodynamics module in order to improve the aerodynamic coefficients based on local geometry variables and stealth constraints. Then, the results are sent to the structural module so that the weight can be minimized against several strength constraints. Finally, the obtained aerodynamic and structural data are send to the global optimizer which evaluates the performance of the aircraft given several mission objectives. According to the authors, the ML strategy can take advantage of different algorithms for different

(35)

processes and it is easier to tackle with because of the considerably fewer global parameters. However, they note that the multiple optimization processes require an increased number of evaluations which can be an additional computational expense.

A typical SL strategy is applied in the case-study of Hitzel et al. [19] for the optimization of UAV configurations. First, a geometry layout is defined in a CAD model which is then analyzed by an aerodynamics model. The obtained results are then used for structural, stability, and propulsion simulations which finally give the necessary data for performance evaluations.

Lundström et al. [15] as well as Amadori et al. [29] also used a SL strategy but the optimization was performed in two successive phases. In the initial phase the calculations were based on simplified equations which allowed a fast evaluation of the optimum propulsion system. After having narrowed down the possible propulsion options and design variables, the second phase was initiated but this time with higher fidelity tools.

2.4.3 O

PTIMIZATION

A

LGORITHMS

Optimization algorithms are an indispensable part of a MDO framework because they automate the iterative optimization process and hence, they enable the system to search for an optimal solution inside the design space [40]. According to Tarkian [2], the algorithms that are used in numerical methods can be classified into two main groups which are characterized as gradient and non-gradient. Safavi [40] explains that “gradient-based methods are used when the gradient of the function is easily accessible

and calculable”, whereas “non-gradient methods are common in non-differentiable, discrete, non-smooth and non-linear engineering problems”.

An evaluation of both gradient and non-gradient algorithms can be found in the UCAV case-study of Amadori et al. [18]. First, they tested the gradient-based algorithms Fmincon which was found to be very fast but unsuccessful for their application because of its inability to escape the local optimum. As a second attempt, they used the non-gradient algorithm Complex together with “forgetting” and “randomization” functions and got significantly better results but the algorithm did not always reach the absolute optimum. Finally, they tested a Genetic Algorithm (GA) and identified that it requires a longer computing time, but it has an improved hit-rate and higher probability of finding the true optimum solution which makes it far more suitable for complex applications.

The effectiveness of GA in optimization of complex products is also expressed in the case-study of MAV by Lundström et al. [15] as well as Amadori et al. [29]. In this case, the Multi-Objective Genetic Algorithm (MOGA-II) was selected because of its capability to process multiple objectives and because it was provided by modeFRONTIER which the authors used to build their framework. Overall, the authors concluded that the algorithm was successful in identifying the “pareto front” of dominant designs but they also noted that a high computing time was required. Several tests were also performed in order to analyze how the algorithm operates in respect to the different weighted objectives and constraints. Although a set of optimum control parameters was identified, it was proven that a correct problem formulation was far more important for guiding the algorithm to the region of interest.

2.4.4 E

FFICIENT

C

OMPUTING

The overall computational time of a MDO process depends on the number of designs which will be evaluated and the required analysis time per design. High-fidelity models

(36)

can give a better design quality but will also require several hours for a single simulation which might render many MDO problems impractical [2].

The concept of metamodels or surrogate models or Response Surface Models (RSM) is to develop a mathematical approximation which will replace a computationally expensive model [26]. Clearly, a metamodel cannot always be a perfect approximation of the system and therefore, it should always be validated in order to ensure that it is sufficiently accurate for the intended application [26], [40]. According to Tianyuan et al. [26] a simple process of developing a metamodel begins with the setup of a set of cleverly chosen sample designs (input), then the system is simulated in order to get the response (output), and finally the input and the output data are combined by a suitable approximation method.

Since metamodels are only mathematical approximations it is possible that their output might differ from the real model. In order to measure the deviation between the real model and the metamodel, the Root Mean Square Error (RMSE) approach can be used [40]. If each of the estimated variables from the metamodels is Xmeta,i and the

respective output from the real model is Xreal,i,then for n designs the RMSE is defined as

seen in formula (2.23). For applications where different units have to be compared, then the normalized form of the RMSE can be used as shown in formula (2.24).

𝑅𝑀𝑆𝐸 = √∑𝑛𝑖=1(𝑋𝑟𝑒𝑎𝑙,𝑖 − 𝑋𝑚𝑒𝑡𝑎,𝑖)2

𝑛 (2.23)

𝑁𝑅𝑀𝑆𝐸 =𝑅𝑀𝑆𝐸

𝑋𝑟𝑒𝑎𝑙

̅̅̅̅̅̅̅ (2.24)

Tarkian [2] notes that sampling can be a very complicated task especially if the inputs depend on the outputs which are generated by the other models. According to the same author, a common solution to this problem is to use the Direct Sampling (DS) and Indirect Sampling (IDS) methods which can be seen in Figure 2.6. The author concludes that for specific tasks the IDS can be a more effective method as it provides a better mapping of the design space but on the downside, it requires further post-processing in order to remove samples that are too close to each other.

Figure 2.6 The DS method (left) and the IDS method (right), adapted from Tarkian [2]

Metamodels are often encountered in MDO frameworks of complex engineering products like aircraft as an alternative to computationally expensive models such as CFD or CAD. The work of Tianyuan et al. [26] as well as Choi et al. [16] illustrated that reasonable results can be obtained if metamodels are implemented in the MDO process. In the former case [26], the authors decided to use the “Kriging” model to replace the computationally expensive models which were used for the RCS and CFD calculations.

(37)

The obtained results from the real- and meta-models were evaluated and the output in both cases was found to be in good agreement. A more thorough investigation and implementation of metamodels is presented in the case-study of Safavi et al. [35]. In this case, the authors used modeFRONTIER to create metamodels of all the simulation models of an aircraft MDO framework by using the “Anisotropic Kriging” algorithm and the “Uniform Latin Hypercube (ULH)” sampling method. The deviation from the real models was found to be very low and the authors reported that metamodels can be an efficient and viable alternative for MDO applications of complex nature.

(38)

(39)

C

ONTRIBUTIONS

Part 3 of the thesis includes a detailed analysis of the main framework as well as a description of the module integration and development. Furthermore, the individual model specifications are elaborated and the miscellaneous evaluation results are presented.

3.1 F

RAMEWORK

E

VALUATION

3.1.1 O

VERVIEW

The proposed multidisciplinary framework is aimed towards the development of a UAV aircraft by considering several engineering disciplines such as geometry, aerodynamics, antenna placement, radar signature and mission performance. In order to effectively integrate the aforementioned disciplines, the framework has been further divided into different modules which utilize specific computational tools and execute predefined calculation routines.

The overall development approach is to use low- to medium-fidelity tools at the initial phase of the optimization process so that some of the calculations can be performed relatively fast at a local level and therefore, reduce the number of design variables that are included in the global optimization scheme. The aforementioned local processes are in principle iterative and they are able to define several of the design characteristics which are needed for later calculations in the framework. Hence, it can be said that this approach aims at reducing the design space at a minimum computational cost as it was also shown in the work of Amadori [29] and Lundström [15]. High-fidelity tools are also included in the basic framework structure but they are engaged at a later phase of the optimization process so that a more accurate result can be obtained. A graphical illustration of the inter-disciplinary relations between the modules can be seen in Figure 3.1 and it is analyzed in detail in section 3.2.

The integration methodology which was applied in the present case-study was to prepare the framework in successive steps by developing one module at a time. The division of the framework into modules was further deemed necessary for the following reasons:

 The different tools which are required for a specific operation are grouped together in a local integration platform and hence, the complexity of the main framework is reduced.

 A future refinement within the modules can be done in an efficient way since this will only induce limited local changes instead of the complex and sometimes radical global changes.

 It is possible to optimize each module locally depending on the chosen optimization strategy or the available computational resources.

 The modules can be simulated individually in order to create metamodels and they also provide a platform for simple metamodel implementation.

 Every module can be tested alone without the need of running time consuming iterations of the whole framework.

Development of a Multidisciplinary Design Optimization Framework Applied on UAV Design

Development of a Multidisciplinary

Design Optimization Framework

applied on UAV Design

Athanasios Papageorgiou

Development of a Multidisciplinary

Design Optimization Framework

applied on UAV Design

Athanasios Papageorgiou

UPPHOVSRÄTT

C

OPYRIGHT

ABSTRACT

ACKNOWLEDGMENTS

ABBREVIATIONS

CONTENTS

LIST OF FIGURES

LIST OF TABLES

I

NTRODUCTION

1.1

B

ACKGROUND

1.2

T

HE

IMPO

Z

P

ROJECT

1.2.1

I

G

1.2.2

S

’

P

1.3

M

ETHODOLOGY

1.3.1

O

1.3.2

D

O

T

T

1.3.3

C

A

L

1.3.4

R

Q

1.4

O

BJECTIVES

A

ND

G

OALS

1.5

R

EPORT

O

UTLINE

F

RAME

O

F

R

EFERENCE

2.1

A

IRCRAFT

D

ESIGN

2.1.1

W

A