ROBUST DESIGN OF UNMANNED AERIAL

VEHICLE
Mr. Gaurav Gowlikar Mr Sathish Nallagoni,
Mr.D.ChandraShekar AsstProf
Department of Aeronautical Engineering, Gurunanak Institution Technical Campus, Ibrahimpatnam, R.R. District, Hyderabad,
Telangana,. India.
Email id: gaurav52535@gmail.com, satish.nallagoni9@gmail.com,
chan.iisc@yahoo.com
ABSTRACT:
Aircraft
design
is
a
complex
multidisciplinary problem involving many disciplines e.g.
aerodynamics, propulsion, structure and flight controls.
Such designs problem often has large dimensional design
parameter consisting of non-continuous functions, making
it extremely hard to solve.

calculations. Better systems can now be designed by

analysing and optimizing various options in a short time. This
is highly desirable because better designed systems cost less,
have more capability, and are easy to maintain and operate.
The design of systems can be formulated as problems of
optimization in which a measure of performance is to be
optimized while satisfying all constraints. Many numerical
methods of optimization have been developed and used to
design better systems. This text describes the basic concepts
of optimization methods and their applications to the design
of engineering systems. Design process is emphasized rather
than optimization theory. Various theorems are stated as
results without rigorous proofs. Optimization theory,
numerical methods, and modern computer hardware and
software can be used as tools to design better engineering
systems. The text emphasizes this theme throughout.

The purpose of this project is to design a program to

generate the conceptual design of fixed wing unmanned
air vehicle (UAV) using robust design optimization.
This optimization program written for the project, seek to
optimize dimensions and performance aircraft based on
the desired flight characteristic
1.1. OBJECTIVE
The objective for the project is to design, implement and test a
computer optimization program for the conceptual design of
fixed wing uavs. An optimization program will be designed to
incorporate multiple disciplines such as aerodynamics,
structures, propulsion and flight control parameters to
achieve required performance metrics for the aircraft. The
optimization program will be validated with the fabrication of
the generated design. Feasibility of the optimization program
will be analysis by means of comparison with conventional
aircraft designs, computation simulations and flight data.

Design is an iterative process. The designers experience,

intuition, and ingenuity are required in the design of systems
in most fields of engineering (aerospace, automotive, civil,
chemical, industrial, electrical, mechanical, hydraulic, and
transportation). Iterative implies analysing several trial
designs one after another until an acceptable design is
obtained. The concept of trial designs is important to
understand. In the design process, the designer teams to
complete the project. For most applications the entire design
project must be broken down into several sub problems which
are then treated independently. Each of the sub problems can
be posed as a problem of optimum design. The design process
should be a well-organized activity. To discuss it, we consider
a system evolution model shown in Fig. The process begins
with the identification of a need which may be conceived by
engineers or non-engineers.

1.2 DESIGN PROCESS

The design of complex systems requires data processing and a
large number of calculations. In the recent past, a revolution
in computer technology and numerical computations has
taken place. Todays computers can perform complex
calculations and process large amounts of data rapidly. The
engineering design and optimization processes benefit greatly
from this revolution because they require a large number of

At some stages in the design process, it may appear that the

process can be completely automated, that the designer can be
eliminated from the loop, and that optimization methods and
programs can be used as black boxes. This may be true in
some cases. However, the design of a system is a creative
process

constraints with respect to the independent design

variables or by performing a very large number of
design evaluations. For the former, because of the
large number of design variables and constraints, the
number of gradients required can become quite large.
All these gradients must be recalculated for each
improved design until the optimum is attained, making
calculations tedious and expensive. Such methods also
depend hugely on the functions involved to be
continuous
and
differentiable.
With
the
multidisciplinary nature of an aircraft design problem,
the objective and constraint functions are often
complex and non-differentiable. As such, the use of
other more effective search algorithms must be
considered.

that can be quite complex. It may be ill-defined and a

solution to the design problem may not exist. Problem
functions may not be defined in certain regions of the
design space. Thus, for most practical problems,
designers play a key role in guiding the process to
acceptable regions. Designers must be an integral part
of the process and use their intuition and judgment in
obtaining the final design.
1.3. DIFFERENCE BETWEEN ENGINEERING
ANALYSIS AND DESIGN ACTIVITIES
It is important to recognize differences between
engineering analysis and design activities. The
analysis problem is concerned with determining the
behaviour of an existing system or a trial system being
designed for a given task. Determination of the
behaviour of the system implies calculation of its
response under specified inputs. Therefore, the sizes of
various parts and their configurations are given for the
analysis problem, i.e., the design of the system is
known. On the other hand, the design process
calculates the sizes and shapes of various parts of the
system to meet performance requirements. The design
of a system is a trial and error procedure. We estimate
a design and analyze it to see if it performs according
to given specifications. If it does, we have an
acceptable (feasible) design, although we may still
want to change it to improve its performance. If the
trial design does not work, we need to change it to
come up with an acceptable system. In both cases, we
must be able to analyze designs to make further
decisions. Thus, analysis capability must be available
in the design process.

1.5. UANMANNED AERIAL VEHICLE

The idea of a pilotless aircraft is not a new concept.
The concept of drones
dates back to the mid1800s,When Austrians sent off unmanned, bomb-filled
balloons as a way to attack Venice The drone we see
today started innovation in the early 1900s, and was
originally used for target practice to train military
personnel. It continued to be developed during World
War I, when the Dayton-Wright .
Unmanned Aerial Vehicles (UAVs) have proven their
usefulness in military reconnaissance in recent military
conflicts. Their practical applications have been
expanding to more than military uses. Various sizes of
UAVs are designed to different levels of performance
depending on their application.
UAVs can be
categorized into four different groups: large, medium,
small, and micro as shown in Figure 1.1 .

1.4. LITERATURE REVIEW

Optimization problems in many industrial applications
are extremely hard to solve due to the high level
integration of systems and the increase in
complexity of analysis and design methods for the
evaluation of system performance. Such problems are
characterized by multi-disciplinary model, objective
functions and constraints that are difficult to evaluate
and, in addition, they often have large- dimensional
design parameter spaces that further complicate the
solution of the problem. Good examples of such
problems can be found in the design of aircraft
systems, where many discipline e.g. aerodynamics,
structures, propulsion and flight control, interact in a
complex multidisciplinary analysis, and where the
models of each of the disciplines may be complex
and even discontinuous. The presently known
optimization methods for such problem involve
analysing the gradients of the objective function and

Figure 1.1 UAVs can be divided into four groups by

respect to its sizes and weights.
CHAPTER 2
2.0. ROBUST DESIGN

In todays automotive industry where simulation and

virtual prototyping are increasingly used to reduce
time to market, the design process has changed
considerably since Taguchi invented the quality
planning concept. He found that it is often more costly
to control the causes of manufacturing variation than
making a process insensitive to these variations. By
using simple experimental designs and loss functions,
he often succeeded in greatly improving product
performance by building in quality, in other words,
implementing the quality-by-design idea. Robustness
analysis aims at providing an accurate estimation of
the sensitivity of outputs to the variability on the
inputs, described in terms of random variables
characterized with probabilistic distributions. In
general, standard deviation is used as a measure for the
robustness of the outputs: the smaller the output
standard deviation, the more robust the output.

applications where optimal control concepts are

needed. In addition, there are some applications in
which both optimum design and optimal control
concepts must be used. Sample applications include
robotics and aerospace structures. In this text, optimal
control problems and methods are not described in
detail. It turns out that optimal control problems can be
transformed into optimum design problems and treated
by the methods described in the text. Thus, methods of
optimum design are very powerful and should be
clearly understood.
The objective for the project is to design, implement
and test a computer optimization program for the
conceptual design of fixed wing UAVs. An
optimization program will be designed to incorporate
multiple disciplines such as aerodynamics, structures,
propulsion and flight control parameters to
achieve required performance metrics for the aircraft.
The optimization program will be validated with the
fabrication of the generated design. Feasibility of the
optimization program will be analysis by means of
comparison with conventional aircraft designs,
computation simulations and flight data.

The core ideas Taguchi elaborated on are still valid,

although in a different design context. Today, product
designers increasingly experiment using computer
models instead of physical prototypes. These
experiments typically involved Monte Carlo
simulation, although robustness is suitable as a
criterion for evaluating discrete-event simulation
systems as well .Using simulation instead of
experimenting with physical prototypes yields system
performance improvements and cost savings, only
when design robustness is evaluated properly. As a
robust design optimization software solution,
Optimum incorporates the Taguchi method for robust
design to deliver designs that are more robust with
respect to manufacturing and geometric tolerances.

Optimization problems in many industrial

applications are extremely hard to solve due to the
high level integration of systems and the increase
in complexity of analysis and design methods for the
evaluation of system performance. Such problems are
characterized by multi-disciplinary model, objective
functions and constraints that are difficult to evaluate
and, in addition, they often have large- dimensional
design parameter spaces that further complicate the
solution of the problem. Good examples of such
problems can be found in the design of aircraft
systems, where many disciplines e.g. aerodynamics,
structures, propulsion and flight control, interact in a
complex multidisciplinary analysis, and where the
models of each of the disciplines may be complex
and even discontinuous. The presently known
optimization methods for such problem involve
analyzing the gradients of the objective function and
constraints with respect to the independent design
variables or by performing a very large number of
design evaluations (direct search methods, such as
genetic algorithms). For the former, because of the
large number of design variables and constraints, the
number of gradients required can become quite large.
All these gradients must be recalculated for each
improved design until the optimum is attained,
making calculations tedious and expensive. Such
methods also depend hugely on the functions
involved to be continuous and differentiable. With the
multidisciplinary nature of an aircraft design problem,
the objective and constraint functions are often
complex and non-differentiable. As such, the use of
other more effective search algorithms must be
considered.

Figure 2.1 robust design cycle

2.1. OPTIMUM DESIGN

Optimum design and optimal control of systems are

two separate activities. There are numerous
applications in which methods of optimum design are
useful in designing systems. There are many other

Robustness-based Design Optimization

Monte Carlo simulation furnishes the decision-maker

with a range of possible outcomes and the probabilities
they will occur for any choice of action it shows the
extreme possibilities the outcomes of going for broke
and for the most conservative decision along with all
possible
consequences
for
middle-of-the-road
decisions.

Design engineering involves designing and developing

technology products. Such products are built to fulfill
a well-specified function under more or less welldefined conditions of utilization. In particular in the
automotive industry, vehicles systematically increase
in complexity, requiring thorough design analysis. This
process consists of a number of analysis verifications
on a virtual product. In this stage, the voice of the
customer needs to be reflected as measurable
objectives in terms of functional performance, and
procedures need to be put in place to set performance
targets, based on experience, competitive analysis, and
legislation.

How Monte Carlo simulation works

Monte Carlo simulation performs risk analysis by
building models of possible results by substituting a
range of values a probability distribution for any factor
that has inherent uncertainty. It then calculates results
over and over, each time using a different set of
random values from the probability functions.
Depending upon the number of uncertainties and the
ranges specified for them, a Monte Carlo simulation
could involve thousands or tens of thousands of
recalculations before it is complete. Monte Carlo
simulation produces distributions of possible outcome
values.

Two new optimization paradigms can be identified,

depending on the application area previously
described here. These are called Robustness-based
Design Optimization and Reliability-based Design
Optimization.

By using probability distributions, variables can have

different probabilities of different outcomes
occurring. Probability distributions are a much more
realistic way of describing uncertainty in variables of a
risk analysis During a Monte Carlo simulation, values
are sampled at random from the input probability
distributions. Each set of samples is called
an iteration, and the resulting outcome from that
sample is recorded. Monte Carlo simulation does this
hundreds or thousands of times, and the result is a
probability distribution of possible outcomes. In this
way, Monte Carlo simulation provides a much more
comprehensive view of what may happen. It tells you
not only what could happen, but how likely it is to
happen.

The first optimization method is Robustness-based

Design Optimization. In this optimization scheme, a
measure of the robustness of the system or component
is used as optimization constraint or objective, in order
to meet not the best performance possible, but the best
robust performance possible. This optimal condition is
robust to small design changes so that if small
perturbations of the design parameters happen, the
performance of the system or component doesn't
decrease below the desired quality level, as illustrated
in the graph below.In
Optimus, all robustness
assessment methods (Monte Carlo, FOSM, etc.) can be
used for optimization purposes, thus including robust
design capability as an integral part of a seamless
integrated simulation process.

Monte Carlo simulation provides a number of

advantages
over deterministic, or
single-point
estimate analysis:Probabilistic Results. Results show
not only what could happen, but how likely each
outcome is Graphical Results. Because of the data a
Monte Carlo simulation generates, its easy to create
graphs of different outcomes and their chances of
occurrence. This is important for communicating
findings to other stakeholders.

2.2 MONTE CARLO SIMULATION

Monte Carlo simulation is a computerized
mathematical technique that allows people to account
for risk in quantitative analysis and decision making.
The technique is used by professionals in such widely
disparate fields as finance, project management,
energy, manufacturing, engineering, research and
development, insurance, oil & gas, transportation, and
the environment.

2.3 POLYNOMIAL CHAOS FUNCTION

Earlier, we explored uncertainty analysis using
Monte Carlo simulation methods and we saw how
extensive sampling of the joint pdf of the inputs and
computation of the outputs at all these points is
required. A major limitation of Monte Carlo

methods is the relative

(1/

)). This

slow convergence (

means

that

the

In equation , the PC representation is divided

into zeroth-order (the first term), first-order (the
second term), second-order (the third term) and
third-order (the fourth term) Hermite polynomials
with number of sums corresponding to polynomial
order. In practice, it is notationally more
convenient to have a single sum instead of
summation over multiple indices. Also note that
equation has an infinite number of terms and
hence to implement this expansion in a computer
program we need to truncate at an appropriately
chosen expansion order or number of expansion
terms.

run-time

associated with Monte Carlo methods can be

prohibitive
for
computationally
expensive
functions. We later studied surrogate modeling
techniques as one possible avenue for reducing the
computational effort. In this chapter, we will study
the application of Polynomial Chaos (PC)
expansion schemes to uncertainty analysis. As we
shall see later, PC expansion schemes are
essentially generalized linear models where the
basic functions are specially tailored orthogonal
polynomials.

We now clarify the notation used in PC

expansion by focusing on a two dimensional PC
representation (i.e., an expansion in terms of two

PC expansions provide a powerful tool for arriving

at
computationally tractable representations
of
random variables or stochastic processes.1 In PC
expansions, the stochastic process/random variable
under consideration is represented in a series that is a
generalization of a Fourier series expansion. This idea
was original1y introduced by Norbert Wiener in 1938,
where he used multidimensional random Hermite
polynomials as basic functions to represent stochastic
processes. The basic idea is to project the random
variable (or stochastic process) under consideration
onto a complete set of orthogonal random polynomials
and in this way obtained an efficient and compact
representation that can be easily analyzed. This idea is
how being extensively studied by many researches in
the context of developing numerical schemes for
solving stochastic partial differential equations in
many fields including elasticity, heat transfer,
electromagnetic, biomedical engineering and fluid
flow problems.
Let Y( ),

random variables)

Y( )=

y 0 + y 1 1+ y 2 2 + y 3 ( 121 ) + y 4 ( 1 2) + y 5(221)
=

y 0 + y 1 1 ()+ y 2 2 ()+ y 3 3 ()+ y 4 4 ( ) + y 5 5 ()

Hence,

Y( )=
where

y i t2 H 2 (i ,t )
1

H n (i , , i )

Where

vector

whose

ex p an si on

coefficient

we can

i=0

are multidimensional

Hermite

the following

1. 0=1 i () =0 for i>0 .

2.

components are uncorrelated Gaussian random

variables with zero mean and unit variance. The
PC

i ()

y 1 i ()

The PC basis functions h a v e

properties.

denotes a multivariate

=(i i )

= 1.

+.,

Hermite polynomial of order n in terms of the

random

of Y(

we defined the first PC basis function as

i1

t 1=1 t 2=1

expansion

polynomials
(also referred to as PC basis
functions)
in terms
of the vector whose
components are uncorrelated Gaussian random
variables with zero mean and unit variance. Note
that for notational convenience,

PC

Y( )= Y 0 H 0+ y i H 1( i ) +
1

for the

truncated at P number of terms,

compactly rewrite equation as follow:

denotes the sample space, denote a second-order

stochastic process.3
Wiener showed that any
second-order stochastic
process admits a PC
expansion of the following form

i1=1

) truncated at

second order, which can be written as follows.

, where

of Y(

i ( ) ,i=0 ,1 , 2 , are orthogonal, i.e.

i j = i2 ij

y i . i 1 are
1

deterministic.

 ij is the Kronecker delta operator and

Where

Where n is dimensionality and p is order of the pc

expansion.

is defined as
CHAPTER 3

f ( ) g () = f ( ) g ( ) p ( ) d .

3.0. PROBLEM

Here p() is the weight function corresponding to

the PC basis, which in this case is the joint pdf
of .
In addition to being orthogonal, a non
truncated set

{ () }i=0

is complete,

needed to prove that any random

3.1. Deterministic optimization formulation:

Figure 3.1 Predator missile profile

a fact

variable

) with arbitrary pdf but finite variance

can be represented in a PC expansion.

The mission profile of Predator A is shown in the

Figure 3.1. The Predator A design optimization is

Y(
i

Table 2.1 pc expansion

In practice we need to find how many terms are

needed in the expansion to achieve a desired
degree of accuracy. Several types of orthogonal
p o l y n o m i a l s c a n be used in a PC expansion
(Debusschere
et al., 2004) and Hermite
polynomials
may not always be the most
efficient - however it serves us well here to
introduce some of the fundamental ideas. In
order t o use the single sum representation
equation.
a set of rules are needed in order
to construct the correct PC basis function

i () for index i.

p=0

P=1

interested student to
2004).

(Debusschere
1

P=3

121

2
1
2

1 3 1

6
2

1 22

8
9

2 11

2
6

2 3 2

this

It can be noted from Table that the number

of terms
in the PC expansion increase
significantly with polynomial order. The
total number of terms in a expansion is
given by P + 1 (counting the zeroth-order
term as well) and can be calculated as

1
1

et al.,

technicality here and consider two dimensional

Hermite polynomials and provide them and
their variances explicit up to fourth order .

1 2

we refer the

We omit H n (i , , i )

221

of one-dimensional Hermite Polynomials and

for details on how to construct the PC basis
functions for general dimensionality
and
order, keeping a one-to-one correspondence

i () ,

P=2

and

i () are products

between

(
2

10

P=4

14 6 12+ 3

11

24
6

1 23 1 2

12
13

12 22 1222 +1

14

6
6
24

1 2 3 1 2
24 6 22+ 3

( P+1 )=

(n+ p)!
n! p ! ,

performed to increase the endurance time at the cruise

flight condition with a full payload condition.
i) The climb and ingress is designed for 5 hours.
ii) Egress and descent/land stage is designed for 6
hours.

iii) Loiter time condition is designed for as 1 hour.

other mission. Therefore, it is selected as an objective

function to maximize for UAV

The design condition is considered at cruise condition

with no payload and full payload condition 18 and 24
hours, respectively. The designed condition is a cruise
condition with full payload aimed for Predator A. The
18 design variables including wing, horizontal, and
vertical tail geometry, and flight condition are listed in
Table below after a sensitivity analysis step. The
endurance of UAVs is a main factor for surveillance
and Design variables for multidisciplinary UAV design
optimization.

Maximum endurance =Endur (

Subjected to:

lower bounds

The altitude (h) and flight speed (v) are considered as

noise factor for UAV cruise design condition. The
adjusted variance function is calculated by
implementing a central difference method as following
equation.

upper bounds

units

Wing span
m

14.8

10

20

Wing root chord

m

1.24

1.0

1.4

Wing tip chord

m

0.5

0.3

0.7

Wing sweep
deg.

10

Wing dihedral
deg.

Wing X location
m

3.59

HT span
m

3.5

4.5

HT root chord
m

0.742

0.4

HT tip chord
m

0.742

0.4

HT sweep
deg.

10

HT X location
m

6.28

5.8

7.8

VT span
m

1.14

0.7

1.5

VT root chord
m

0.742

0.4

VT tip chord
m

0.742

0.4

VT LE sweep
deg.

60

VT X location
m

6.82

5.8

7.8

V design
m/s

42

h altitude
m

3000

27.72( V stall)
2000

( Gi ( x ) ) > 0, i=1, ..., 17

3.2. RDO formulation:

Table 3.1 Design variables for multidisciplinary UAV

design optimization
Baseline

x i ) j=1,... 18

64
4000

[(

)] ( )

2
new 2 objective
function for RDO is defined by

)(

f ( v+ v ) f ( v v )
f ( h+ h ) f (hequation
h)
1 2
+
with
1 altitude and speed noise factor
2

v
2

h
consideration
3
in
equation above.
^y 2=
T var
Were h and v are flight speed and altitude step A
A new objective function for RDO is defined by
equation
with altitude and speed noise factor
consideration in equation above.

over drag ratio are satisfied for design requirements.

The UAV gross, empty, and wing weight are satisfied
for constraints which are set by following a baseline
value.
The probabilistic analysis is performed on
deterministic and RDO results with considering flight
speed and altitude as noise factor. The probabilistic
analysis results are shown in Figure and Table below.
The 1500 sampling points are used for MCS, then the
distribution are fitted as shown in Figure. The reasons
are a central difference method directly used into
establishing a new RDO objective which requires a
four times running Advanced UAV to complete one
iteration. In addition, MCS is normally required at
least on millions sampling points to obtain a
simulation results. Therefore, the fitting method is
applied to generate a fit distribution curve in Figure
11 while still maintaining an accuracy of probabilistic
analysis results and saving MCS run. The target
endurance mean value and target variance value is set
21 hours and 1.5 hours respectively. The comparison
are presented in Table with a small improvement in
an endurance objective function at 3.46%. However,
the variance reduces from 9.57 hours to 1.5 hours
with 84.33% improvement which is also seen by
Figure 11 while the RDO fitting distribution provides
a narrow and more reliable results comparing with a
deterministic UAV results. The reliability is also
increased by 55.56% comparing with a deterministic
results due to RDO process that reduces less
sensitivities of noise factor to an endurance objective
function. Therefore, RDO process demonstrates the
improvement of reliability UAV performance results
while considering uncertainty parameters and
provides trustworthy design results comparing with
a traditional deterministic design optimization
approach at the conceptual design stage.

CHAPTER 4
4.0 Deterministic and robust design optimum
results
The deterministic and robust design optimum
configuration results for multidisciplinary UAV
design optimization are presented in Table below
comparing with a baseline - Predator A configuration.
The endurance objective function shows an
improvement in both deterministic and RDO UAV
results from 19.23 hours to 21.13 and 22.26 hours for
deterministic and robust design results respectively by
the helps of Design Explorer algorithm. The objective
convergent history and maximum violation constraints
for a deterministic formulation are shown in Figure 9.
The endurance objective function converges at 21.13
hours with 1% constraints violation in Figure b.
The new objective function for RDO includes an
adjusted endurance mean function and adjusted
variance which converges to 1 with no constraints
violation shown in Figure 10a and 10b. The wing
weight and directional derivatives coefficient
constraints are active. Hence, the design formulation
problem for deterministic and robust design process
are converged and strictly formulated. The maximum
constraints violation graphs are also shown that if
designers are accepted for more risk in constraints
violation, the endurance objective value is also
increased.
The horizontal tail root chord and sweep angle
variables are hit upper bound for robust design results
due to increment in wing area and wing sweep angle
and reduction in wing location. Hence, the horizontal
tail is required to increase a span to satisfy static
stability conditions such as static margin and pitching
moment coefficient. The vertical tail geometry and its
location optimum results are also satisfied the
directional and lateral constraints as shown in Table
6. The optimum results are also recommended to
flight at lower speed and higher altitude to improve
an endurance as shown in RDO UAV results column.
Other performance constraints such as takeoff,
landing distance, stall speed, maximum speed, and lift

The computational strategy for a proposed robust

UAV design implements a central difference method
and fitting distribution for 1500 sampling points of
MCS which shows extremely computational time
savings comparing with normal RDO method required
at least 1 million sampling points shown in Table. The
computer performance configuration is i7, 3.07 GHz
and 16GB of RAM with a parallel computing set for
4 CPUs. The proposed UAV RDO framework is only
required almost 21 hours to obtain a probabilistic
analysis results comparing

Baseline

Objective function

Endurance

UAVRDO UAV result

Unit

19.23

21.13

22.26

hours

Wing span

14.8

15.01

13.2

Wing root chord

1.24

1.23

1.31

Wing tip chord

0.5

0.52

0.67

Wing sweep

7.85

7.97

deg.

Wing dihedral

1.5

deg.

Wing location

3.59

3.43

3.28

3.51

3.95

HT span
HT root chord

0.742

0.743

HT tip chord

0.742

0.695

0.56

10

10

HT location

6.82

6.24

6.78

VT span

1.14

1.6

1.48

HT sweep

Design Variables

Deterministic
result

deg.

VT tip chord

0.742

0.958

0.84

VT root chord

0.742

0.77

0.84

6.44

3.75

deg.

6.82

7.8

6.38

42

38

32.1

m/s

3000

2754

3793

Wing taper ratio

0.4

0.42

0.511

Lift to drag ratio

21.3

21.55

22.11

Maximum speed

64.35

64

66

m/s

Stall speed

24.48

24.27

24.23

m/s

Takeoff ground roll

387.6

381

Takeoff field length

651.7

647.6

651.7

Landing distance

402.7

397.8

395.6

VT LE sweep
VT location
Vdesign
Flight altitude

377.95

Clb

0.0069

-0.123

Cnb

0.0055

0.05

0.049

-1.074

-1.46

-1.14

Static Margin

0.156

0.15

0.179

Wing weight

91

91

91

kg

Cm

Constraints

MTOW
Empty weight

-0.208

1011.7

1010

1006.4

kg

334.7

333.2

329.34

kg

Table 4.1 UAVs deterministic and RDO design results comparison

2
2

Deterministic
results

Endurance

21.11

Variance

9.5

Improve
ment
(%)

Unit

21.84

3.46

Hrs.

1.5

84.33

Hrs.

Robust design
results

Dimitri N Mavris, Oliver Bandte, and Daniel A

DeLaurentis, Robust Design Simulation: A
Probabilistic Approach to Multidisciplinary Design,
Journal of Aircraft, vol. 36, no. 1, pp. 298-307, 1999.

Table 4.3 Computational strategy of probabilistic

analysis for UAV robust framework
comparison

No.
of
calls:
Advanced UAV
Execution time (s):
Advanced UAV
Monte
Carlo
Simulation (MCS)
Total computation
time (hours)
Total computation
(days)

UAV
framework
5

40

40

100000

1500

2778

20.75

116

3/4

REFERENCE:
J. Sobieszczanski-Sobieski and R. T. Haftka,
Multidisciplinary aerospace design optimization:
survey of recent developments, Structural and
Multidisciplinary Optimization, vol. 14, no. 1, pp. 123, 1997.

Table 4.2 Deterministic and robust design

results probabilistic analysis comparison

Normal RDO

provides a robust optimum UAV configuration

without any noticeable increase in design turnaround
time under multidisciplinary design optimization
environment

Gyung-Jin Park, Tae-Hee Lee, Kwon Hee Lee, and

Kwang-Hyeon Hwang, Robust Design: An Overview,
AIAA Journal, vol. 44, no. 1, pp. 181-191, 2006.

RDO

I. Doltsinis and Z. Kang, Robust design of structures

using optimization methods, Computer Methods in
Applied Mechanics and Engineering, vol. 193, no. 23
26, pp. 2221-2237, Jun. 2004.
Hyeong-Uk Park, Sang-Ho Kim, Jae-Woo Lee, and
Yung-Hwan Byun, Design of Very Light Jet (VLJ)
Aircraft Using Robust Design Optimization
Approach, in CJK 5th Proceedings, Jeju, Korea, 2008.

CONCLUSION:

W. Chen, R. Garimella, and N. Michelena, Robust

design for improved vehicle handling under a range
of maneuver conditions, Engineering Optimization,
vol. 33, no. 3, pp. 303-326, Feb. 2001

A multidisciplinary robust optimization framework

for UAV conceptual design is developed and applied
successfully for improving a Predator A endurance
with a trustworthy optimum configuration with a
55.56% improvement in reliability and 84.33%
improvement of variance reduction comparing with a
traditional deterministic approach. The robust
optimization
framework
demonstrates
an
effectiveness of reducing less sensitivity of noise
factors to an endurance objective function in UAV
conceptual design case study. Especially,the
computational strategy for probabilistic analysis
shows extremely computational time saving
comparing with an expected normal RDO approach
with a same computer performance.

Wei Chen and Kemper Lewis, Robust design

approach for achieving flexibility in multidisciplinary
design, AIAA Journal, Vol. 37, No. 8, August 1999
Xiaoping Du and Wei Chen, Efficient uncertainty
analysis methods for multidisciplinary robust design,
AIAA Journal, Vol. 40, No. 3, March 2002.
Ioannis Doltsinis and Zhan Kang, Robust design of
structures
using
optimization
methods,
Computationa Methods Applied Mechanics and
Engineering. 193 (2004) p. 2221-2237, Elsevier
Eric Sandgren and T.M. Cameron, Robust design
optimization of structures through consideration of
variation, Computers and structures 80 (2002) p.
1605-1613, Pergamon, Elsevier

Advanced UAV analysis tool is well-developed for

a multidisciplinary UAV conceptual design and
integrated into a robust optimization framework.
Aerodynamics, propeller, weight, and performance
analysis are validated for a baseline Predator A with a
maximum error at 9.28% which is acceptance for a
conceptual design stage.

ModelCenter
10.1,
PHX
ModelCenter:
http://www.phoenix-int.com/software/phxmodelcenter.php
Hyeong-Uk Park, Jae-Woo Lee, and Yung-Hwan
Byun, Development of the Robust Aerospace

A multidisciplinary UAV design formulation is

successfully made for a robust framework which

10

System Design Process, 21st Canadian Congress of

Applied Mechanics, Department of Mechanical &
Industrial Engineering, Ryerson University, Toronto,
Ontario, CANADA, 3-7 June 2007

Nhu-Van Nguyen, Seok-Min Choi, Wan-Sub Kim,

Sangho Kim, Jae-Woo Lee, and Yung-Hwan Byun,
Multidisciplinary Unmanned Combat Air Vehicle
(UCAV) System Design Implementing MultiFidelity Models, Aerospace Science & Technology,
Vol. 26, Issue 1, p. 200-210, 2012, Elsevier
Publisher.

Abdulaziz Azamatov, Jae-Woo Lee, and YungHwan Byun, Comprehensive aircraft configuration
design tool for Integrated Product and Process
Development, Advances in Engineering Software,
vol. 42, no. 12, Jan. 2011.

Andrew J. Booker, Martin Meckesheimer, and Tony

Torng, Reliability Based Design Optimization Using
Design Explorer, Optimization and Engineering,
June 2004, Volume 5, Issue 2, pp 179-205

Nhu-Van Nguyen, An Efficient Multi-Fidelity

Approach for the Multi-Disciplinary Aerospace
System Design and Optimization, Ph.D thesis,
Konkuk University, Seoul, Korea, 2011.

J. F. Gundlach IV, Multi-disciplinary design

optimization of subsonic fixed-wing unmanned aerial
vehicles projected through 2025, Virginia
Polytechnic Institute and State University, 2004.

11