Professional Documents
Culture Documents
net/publication/268477521
CITATIONS READS
3 140
4 authors, including:
Marin Guenov
Cranfield University
69 PUBLICATIONS 872 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Marin Guenov on 09 May 2016.
[Abstract] The objective of this paper is to present a novel methodology for enabling the
visualization and exploration of data obtained during aircraft multiobjective optimization at
conceptual design stage. The aim is to enhance the designer’s insight into the optimization
problem at hand. To achieve this goal, the present methodology integrates suitable means for
visualization and exploration of the optimal solutions. By combining different graphical
interfaces, the designer is provided with diverse perspectives of the multidimensional data
under study. Additionally, the method provides a posteriori exploration of further optimal
solutions. This is made possible by means of appropriate approximation techniques which
are integrated within the visualization environment and allow obtaining such information at
no additional computational cost. The application of the present methodology to an aircraft
sizing test case shows promising results in helping the designer understand the complex
high-dimensional data involved.
Nomenclature
1
American Institute of Aeronautics and Astronautics
S = aircraft reference area
SMP = scatter matrix plot
SOM = self organizing map
T = aircraft thrust
V = aircraft speed
W = aircraft weight
x = design vector
xi = i-th design variable
I. Introduction
A ircraft multidisciplinary design optimization (MDO) takes into account simultaneously the contributions of
different disciplines, which often exhibit strong coupling. Most of the MDO research effort so far has been
focused on handling the interactions between analysis tools, including the development of novel numerical
techniques. However, the large multidimensional datasets resulting from such an approach are often too complex to
be analyzed and completely understood by the designers who are accustomed to traditional design visualization
tools. There is an apparent need for a suitable visualization methodology to make the results of a complex design
optimization procedure fully readable and meaningful to the designer. The development of such a methodology is
the subject of this paper.
The next section presents the formulation of the optimization problem and reviews multidimensional
visualization techniques used in the field. Section III presents the novel integrated visualization methodology.
Section IV focuses on the application of the method to a mid-range civil aircraft sizing test case. Finally,
conclusions are drawn in section V.
II. Background
A. Optimization
The aim of conceptual design is to develop the overall aircraft configuration arrangement for a given set of
requirements and specifications. Since a large number of alternative design concepts are evaluated, it follows that
during this phase an iterative process, including continuous analysis, changes and improvements of the design
layouts, is required to achieve a consistent design. As the design solutions advance step by step towards more
detailed design, it becomes more difficult and expensive to introduce changes. Therefore, it is fundamental to obtain
a good design at conceptual stage, so that the following design phases will not require major revisions. In this
context the design of an aircraft frequently results in optimizing multiple criteria for a large number of design
variables with the requirement to meet a set of constraints. For example, the designer may wish to maximize the
aircraft range and minimize its maximum take-off weight according to constraints related to performance
characteristics, regulations, safety and maintenance considerations.
Such design problems, referred to as multiobjective optimization (MOO) problems, are very complex and
require specific numerical methods to obtain solutions.
min F (x) =
x∈S
{ f (x), f (x), ... , f (x)}
1 2 M
x = { x , x , ... , x }
1 2 N (1)
where F(x) is the vector of the M objectives to be minimized with respect to the design vector x in the N-
dimensional design space S, subject to L constraints. A constraint gi is said to be active at a point x* of the design
space S if a strict equality holds at this point, that is, gi (x) = 0.
2
American Institute of Aeronautics and Astronautics
2. Pareto solutions and approximation
Solving a MOO where the objectives are conflicting usually does not have a unique solution but a set of non-
dominated solutions also known as Pareto solutions. A feasible design point is said to be Pareto optimal if no other
feasible design can improve some of the objectives without simultaneously being detrimental to others. For real-life
engineering problems, the Pareto frontier cannot be described analytically. Numerical methods are thus required to
obtain discrete Pareto solutions, which turns out to be computationally demanding. Research has been carried out to
locally approximate the Pareto frontier by reusing gradient information obtained during the optimization procedure.
These methods allow to derive new approximate Pareto solutions in the objective space and to obtain their
corresponding design vectors in the design space1,2. These approximations based on Taylor expansion on the Pareto
frontier are defined as follows:
nf
df p
fp = f p* + ∑ df
i =1 i
Δfi ( p = n f + 1, ..., M ) (2)
nf nf
df p 1
f p = f p* + ∑ df
i =1 i
Δf i +
2 ∑H
j , k =1
( p)
jk Δf j Δf k ( p = n f + 1, ..., M ) (3)
df p d2 fp
where and H (jkp ) = are the first and second order derivatives on the Pareto surface, nf is the dimension of
df i df j df k
the largest family of linearly independent vectors P∇fi, where P is the projection matrix onto the hyper-plane
normal to all gradients of active constraints. The derivatives can be computed from the usual gradients of objectives
and active constraints obtained during the optimization process. Thus further information in the vicinity of a Pareto
point can be obtained at no extra computational cost.
In this paper we will refer to the term optimal family of solutions to identify those designs belonging to the same
local Pareto frontier and characterized by the same set of active constraints (including the active bounds of the
variables).
B. Visualization
A major requirement for an effective visualization technique is to be able to translate numerical datasets into
simple and meaningful graphical representations in order to facilitate data analysis and understanding. Previous
efforts in this field have been based on the application of multidimensional visualization techniques in MOO, so that
both evaluation and exploration of the Pareto frontier could be performed. In some cases, well-known methods have
been implemented3,4, while in others ad hoc methodologies have been developed5,6. A brief summary is presented
below.
2. Carpet plots
Prior to the application of computational tools in aircraft conceptual design, optimization methods were based on
the development of a set of parallel layouts, each one characterized by different combinations of the design
3
American Institute of Aeronautics and Astronautics
parameters. Design optimization was carried out with the help of carpet and matrix plots11-13 by estimating the
impact of parametric variations on the aircraft layout and criteria such as mission, weight and cost. A typical carpet
plot provides a means of visualizing performance requirements (e.g. cruise speed, second segment climb rate, take-
off and landing field length distances) as a function of parametric variables such as thrust-to-weight-ratio (T/W) and
wing-loading (W/S). A point on the carpet plot represents a particular aircraft design and provides the designer with
information on how performance constraints are satisfied.
A. Methodology
The proposed approach is aimed at enhancing the visualization of design solutions in an MOO framework. It is
based on the authors’ belief that in order to convey the full meaning of design optimization data to the designer a
combination of visualization techniques have to be used together and also that the interpretation of such data should
not require the designer to be proficient in numerical optimization methods. Depending on the analysis to be carried
out, suitable methods have to be selected with respect to key features of the datasets to be investigated in the attempt
to guarantee their full readability. The present methodology is based on the development of a matrix (Table 1) which
identifies the suitable visualization techniques to be used in the context of common data analysis scenarios occurring
during aircraft design optimization.
Technique
Carpet Plots SMP PCP Objective Space
Scenario
Visualization of a Visualization of input, objective Objective value
Performance analysis
single Pareto point and constraint values comparison
Objective value
Visualization of a
Performance comparison Trend and correlation Identification of design points comparison and
reduced number of
between design points analysis sharing common features visualization of optimal
Pareto points (<10)
families of designs
Objective value
Visualization of a Analysis/Visualization of
Trend and correlation Identification of design points comparison and
large number of the entire solutions-set
analysis sharing common features visualization of optimal
Pareto points (>10) distribution
families of designs
Objective value
Visualization of Performance visualization Gradient information Identification of common
comparison and
Pareto frontier of approximated Pareto for local sensitivity features shared between the
visualization of optimal
approximation frontier analysis approximated solutions
families of designs
Constraint Performance requirements Study of how well constraints Objective value
satisfaction study check are satisfied comparison
Comparison of designs Analysis of designs Visualization of
Active constraints
characterized by constraint characterized by constraint optimal families of
study
activation activation designs
4
American Institute of Aeronautics and Astronautics
A visualization tool implementing this methodology has also been developed: the Integrated Exploration and
Visualization Interface (IEVI). It integrates three graphical interfaces whose synergic coupling aims at providing the
designer with improved insight into the data for various analysis scenarios. The three graphical interfaces are
described in detail in the following sub-sections. Each interface is focused on the representation of a particular
perspective of the aircraft multiobjective optimization at conceptual design stage. The simultaneous update and
interactive application of such interfaces aims to further improve the visualization of the complex information
produced by the design optimization tools. In addition, the method is intended to provide a posteriori exploration of
additional optimal solutions. This is made possible by means of the approximation technique described above, which
is integrated into the visualization environment and do not entail any additional computational cost. A typical use of
the visualization tool is given in Figure 1 while Table 1 identifies the set of suitable visualization techniques to be
used for each common data analysis scenario occurring in the aircraft design optimization (green boxes in Fig. 1).
Figure 1: Example of the procedure for the use of the visualization tool IEVI.
5
American Institute of Aeronautics and Astronautics
B. Visualization Interfaces
- The identification of the objective minimum and maximum values for the set of optimum points;
- The position of local Pareto regions, which correspond to different optimal families of solutions
sharing the same features;
- The density of the design solutions in a specific area of the Pareto front;
- The objective space location of the solutions characterized by one or more specific active constraints.
Figure 2: Example of the carpet plots of two different design points, one in red and the other in blue.
Hatching denotes inadmissible side of constraint curves.
6
American Institute of Aeronautics and Astronautics
This graphical interface gives the designer an effective way to assess any design solution. The integration of
carpet plots in the visualization framework allows the designer to evaluate the Pareto frontier with respect to a set of
design constraints by using a traditional design tool without the burden of mathematical complexity.
As outlined in section II, approximated solutions can be obtained at no extra cost in the vicinity of any Pareto
solution of interest. These approximated solutions can be plotted both in the OSI and CPI which gives the designer
the possibility to explore other Pareto solutions positioned in a more desirable area of the T/W-W/S space. Such a
posteriori analysis allows finding design solutions which are approximately Pareto optimal, but may help meeting
additional criteria not considered in the initial MOO formulation.
- the user can choose the graphical interface to interact with amongst OSI, CPI and MDVI;
- diverse perspectives on the design under consideration can be enabled by a simultaneous update of the
visualization on the remaining interfaces;
The above features are also applicable to multiple design points comparison.
2. Constraint Study
A constraint behavior analysis is required to enhance the designer’s insight into the optimization outcome. This
can be achieved in two different ways:
- the user can click on a design point in the OSI, so that all the constraints are visualized in the PCP and the
SPM. In the former graph, it would be possible to numerically check constraints satisfaction. Moreover, all
active and inactive performance constraints will be represented in different colors, for example, in red and
green, respectively. In the SPM plot, a local sensitivity analysis will be provided by conveying the effect of
each variable on each constraint;
- the decision-maker can select a constraint in the PCP in order to visualize all the solutions for which it is
active in the OSI and CPI.
7
American Institute of Aeronautics and Astronautics
Both approaches allow the user to gain a better understanding of the design and objectives spaces. Depending on
the information the designer is interested in, it may be essential to discern:
- what solutions are characterized by the activation of one or more particular constraints;
- what are the active constraints that feature each optimal family of solutions;
- what constraints have an influence on a specific area of the Pareto frontier or of the Carpet Plot space;
- what variables and what range of values are determinant for the activation of each constraint.
Wing Data:
S Reference wing area [ft2]
TR Taper ratio of the wing
SWEEP Quarter-chord sweep angle of the wing [deg]
TCA Wing thickness-chord ratio
SPAN wing span [ft]
DIH Wing dihedral (positive) or anhedral (negative) [deg]
FCOMP Decimal fraction of amount of composites used in wing structure
Tails, Fins, Canards Data:
SHT Horizontal tail theoretical area [ft2]
SWPHT Horizontal tail 25% chord sweep angle [deg]
ARHT Horizontal tail theoretical aspect ratio
TRHT Horizontal tail theoretical taper ratio
TCHT Thickness-chord ratio for the horizontal tail
Vertical Tail Data:
SVT Vertical tail theoretical area (per tail) [ft2]
SWPVT Vertical tail sweep angle at 25% chord [deg]
ARVT Vertical tail theoretical aspect ratio
TRVT Vertical tail theoretical taper ratio
TCVT Thickness-chord ratio for the vertical tail
Propulsion System Data:
THRSO Rated thrust of baseline engine [lb]
BPR Bypass ratio
NEW Number of wing mounted engines
NEF Number of fuselage mounted engines
Configuration Data:
DESRNG Design range [NM]
MTOW Maximum Take-Off Weight [lb]
VCMN Cruise Mach number
8
American Institute of Aeronautics and Astronautics
CH Maximum cruise altitude [ft]
HTVC Modified horizontal tail volume coefficient
VTVC Modified vertical tail volume coefficient
HHT Decimal fraction of vertical tail span where horizontal tail is mounted
VAPPR Maximum allowable landing approach velocity [kts]
FLTO Maximum allowable takeoff field length [ft]
WFC Wing fuel capacity [lb]
FFC Fuselage fuel capacity [lb]
Crew and Payload Data:
NPF Number of first class passengers
NPB Number of business class passengers
NPT Number of tourist class passengers
NSTU Number of flight attendants
Fuselage Design Data:
FPITCH Seat pitch for the first class passengers [in]
NFABR Number of first class passengers abreast
BPITCH Seat pitch for business class passengers, [in]
NBABR Number of business class passengers abreast
TPITCH Seat pitch for tourist class passengers [in]
NTABR Number of tourist class passengers abreast
Constant Variables
NEW = 2 NEF = 0 NPF = 20
NPB = 20 NPT = 150 NSTU = 10
FPITCH = 34 in NFABR = 4 BPITCH = 30 in
NBABR = 4 TPITCH = 28 in NTABR = 6
HTVC = 1 VTVC = 1 HHT = 0.02
Input Variables x ∈ [xlb,xub]
SPAN = [90,130] ft DIH = [-4,4] deg FCOMP = [0,0.35]
2 SWPHT = [30,32] deg ARHT = [2.5,3.5]
SHT = [200,250] ft
TRHT = [0.225,0.25] TCHT = [0.07,0.1] 2
SVT = [200,250] ft
SWPVT = [30,32] deg ARVT = [1.3,1.7] TRVT = [0.225,0.25]
TCVT = [0.07,0.1] THRSO = [29200,32000] lb DESRNG = [2500,3000] NM
2 TR = [0.225,0.25] SWEEP = [30,32] deg
S = [1500,1800] ft
TCA = [0.07,0.085] VCMN = [0.78,0.82] CH = [25000,35000] ft
BPR = [5,6]
⎧⎪ MTOW ⎫⎪
min F (x) = ⎨ ⎬
x
⎪⎩ Total Fuel ⎭⎪
Subject to:
9
American Institute of Aeronautics and Astronautics
⎛ 1 ⎡W ⎤ ⎞
⎜ C ⋅ ⋅ ρ ⋅V 2 ⎢S⎥ ⎛ W ⎞ ⎟⎟
2
⎡T ⎤ TTO ⎜ D0 2 ⎣ ⎦TO
g 4 ( x) = ⎢ ⎥ − ⋅ + ⋅⎜ ⎟ ≥ 0;
⎣ W ⎦TO T ⎜ ⎡W ⎤ 1
⋅ ρ ⋅ V 2
⋅ π ⋅ A ⋅ e ⎝ WTO ⎠ ⎟
⎜ ⎢S⎥ 2 ⎟
⎝ ⎣ ⎦TO ⎠
with:
x lb ≤ x ≤ x ub ;
where the subscripts TO and 2 refer to the take-off and second segment conditions, the sub-subscript 1e denotes the
thrust of a single engine, the constraints g3 and g4 represent the requirements on second segment with one engine
inoperative and cruise speed respectively.
B. Data Analysis
Solutions of the above MOO problem have been obtained using an optimization tool developed at Cranfield
University14,15. The default visualization mode of the optimization outcome is displayed in Figure 3, where all the
design solutions are displayed simultaneously in each graphical interface. For this graphical mode the CPI is aimed
at highlighting the region of the T/W-W/S space where the Pareto solutions are clustered, while the PCP provides a
numerical data perspective only for the axes required by the user. Through these visual settings the designer is
helped in understanding the overall features of the results set at hand, especially by highlighting the available ranges
of values for each design parameter.
Figure 3: Default visualization of the optimization results on the Integrated Exploration and Visualization
Interface (IEVI)
10
American Institute of Aeronautics and Astronautics
The Pareto exploration and analysis can be conducted by interactively selecting the design of interest on any
graphical interface (OSI, CPI and MDVI) thus updating the other interfaces visualization only with the data of the
sample under study. Alternatively, the user can compare solutions by clicking on a set of Pareto points as it is shown
in Figure 4, where the corresponding carpet plots and polylines are displayed in the CPI and PCP, respectively.
Figure 4: Results visualization in Interactive Mode for OSI and in Multiple Samples Mode for CPI and PCP.
By clicking on the optimal design solutions of interest in the OSI, the user can compare these by analyzing
further data perspectives of the selected samples on the other interfaces, which are updated simultaneously.
For each axis selected in the PCP, the horizontal green ticks show the minimum and maximum values of the set
of Pareto solutions obtained during the optimization. Furthermore, the designer can easily identify all the designs
having similar features with respect to a particular set of parameters. This is made possible by specifying the range
of values to satisfy for the parameters of interest. In Figure 5 the green horizontal ticks identify again the bounds of
the Pareto set for each parameter, whereas the yellow ticks represent the ranges interactively chosen by the user for
three different parameters. It should be noted that in this hypothetic data-analysis case the designer is interested in
identifying all those solutions for which:
11
American Institute of Aeronautics and Astronautics
Figure 5: PCP in Selective Ranges Mode.
Figure 6: Visualization of the Pareto solutions (in green in the OSI and CPI) obtained during the
optimization procedure and the local approximations (in orange) of the design point under study (in sky
blue). The green ticks displayed in the PCP represent the bounds of the Pareto set for each parameter.
12
American Institute of Aeronautics and Astronautics
In Figure 6 and Figure 7 the approximation capabilities of the visualization tool have been used to obtain new
approximated Pareto solutions in an attempt to explore a region of the Pareto front in the vicinity of a particular
solution. The approximation is displayed with a series of orange points both in the OSI and in the CPI. This enables
the designer to assess performance constraint satisfaction and also to understand how a particular area of the Pareto
front is mapped to the (W/S – T/W) space. Additionally, the PCP of the approximated Pareto solutions can be
visualized through a set of polylines of the same color. This provides extra information about the relationships
among the optimization variables and allows the designer to identify key features of the designs in the vicinity of the
solution under consideration. The user can easily assess the range of variations observed on the Pareto front and
make comparisons between the variables and objectives changes derived with the approximation.
These approximation capabilities can be very useful for computationally and time demanding problems. The user
is allowed to reduce the optimization efforts by obtaining only a reduced number of optimal points and,
subsequently, to explore through a set of approximated solutions the Pareto frontier such points belong to.
Figure 7: Zoom-in on the CPI – carpet plot of the Pareto solution under study (sky blue point)
including all Pareto solutions obtained during the optimization procedure (green) and some
approximated Pareto solutions (orange).
Presented is a novel methodology aiming to enhance the decision making process associated with aircraft
multiobjective optimization at conceptual design stage. It includes the integration of suitable visualization and
exploration capabilities. The combination of different graphical interfaces and visualization techniques provides the
designer with diverse perspectives on the data under study and bridges the gap between MOO and the more
conventional design approaches. Ultimately, the method allows the designer to explore further optimal solutions
13
American Institute of Aeronautics and Astronautics
with respect to other criteria not considered in the initial MOO formulation. This is made possible by means of
proper approximation techniques integrated within the visualization environment and without any significant
additional computational cost. The methodology has been tested with a sizing test case in order to evaluate and
validate the method capabilities.
- integrating the SPM in order to allow the user to obtain and display the local sensitivity on the Pareto
surface of the point under study for a chosen set of input variables, constraints and objectives;
- integrating additional multidimensional visualization methods to display Pareto frontiers with more than
three objectives;
- integrating uncertainty visualization techniques in order to facilitate the results analysis of robust
optimization problems.
Acknowledgments
The authors gratefully thank L. Arnold McCullers at NASA Langley for supplying FLOPS.
References
1
Utyuzhnikov, S.V., Maginot, J., and Guenov, M., “Local Pareto Analyzer for Preliminary Design,” Proceedings of the 25th
International Congress of the Aeronautical Sciences, ICAS, Stockholm, Sweden, 2006.
2
Utyuzhnikov, S.V., Maginot, J., and Guenov, M., “Local Approximation of Pareto Surface,” Proceedings of the World
Congress on Engineering, Vol. 2, IAENG, Hong Kong, 2007, pp. 898-903.
3
Holden, C. M. E., and Keane., A. J., “Visualization Methodologies in Aircraft Design,” 10th AIAA/ISSMO Multidisciplinary
Analysis and Optimization Conference, AIAA, Washington, DC, 2004, 2004-4434.
4
Grasmeyer, J., “Multidisciplinary Design Optimization of a Transonic Strut-Braced Wing Aircraft,” 37th AIAA Aerospace
Sciences Meeting and Exhibit, AIAA, Washington, DC, 1999, AIAA 99-0010.
5
Agrawal, G., Lewis, K., Chugh, K., Huang, C.-H., Parashar, S., and Bloebaum, C. L., “Intuitive Visualization of Pareto
Frontier for Multi-Objective Optimization in n-Dimensional Performance Space,” 10th AIAA/ISSMO Multidisciplinary Analysis
and Optimization Conference, AIAA, Washington, DC, 2004, 2004-4434.
6
Deremaux, Y., Willcox, K., and Haimes, R., “Physically-Based, Real-Time Visualization and Constraint Analysis in
Multidisciplinary Design Optimization,” 33rd AIAA Fluid Dynamics Conference and Exhibit, AIAA, Washington, DC, 2003,
2003-3876.
7
Jacoby, W. G., Statistical Graphics for Visualizing Multivariate Data, Sage Publications, 1998.
8
Young, F. W., Valero-Mora, P. M., and Friendly, M., Visual Statistics: Seeing Data with Dynamic Interactive Graphics,
John Wiley, New York, 2006.
9
Kohonen, T., “The Self-Organizing Map,” Proceedings of the IEEE, Vol. 78, No. 9, 1990, pp. 1464-1480.
10
Stump, G. M., Simpson, T. W., Yukish, M., and Bennett, L., “Multidimensional Visualization and Its Application to a
Design by Shopping Paradigm”, 9th AIAA/USAF/NASA/ISSMO Symposium on Multidisciplinary Analysis and Optimization,
AIAA, Washington, DC, 2002, 2002-5622.
11
Raymer, D. P., “Aircraft Design: A Conceptual Approach,” AIAA Education Series, Second Edition, AIAA, New York,
1992.
12
Loftin, L. K. Jr., “Subsonic Aircraft: Evolution and the Matching of Size to Performance,” NASA, RP-1060, 1980.
13
Roskam, J., “Airplane Design: Part I, Preliminary Sizing of Airplanes,” Roskam Aviation and Engineering Corp., Ottawa,
KS, 1979.
14
Fantini, P. “Improving the Effectiveness of Multi-Disciplinary Design Optimization at the Aircraft Conceptual Design
Phase,” Ph.D. Dissertation, Aerospace Engineering Department, Cranfield University, UK, 2007.
15
Fantini, P., Balachandran, L. K., and Guenov, M. D., “Computational Intelligence in Multi-Disciplinary Optimization at
Feasibility Design Stage,” First International Conference on Multidisciplinary Design Optimization and Applications, ASMDO,
2007.
16
McCullers, L. A., FLOPS User’s Guide, Release 7.40. Text file included with the FLOPS code.
14
American Institute of Aeronautics and Astronautics