Aircraft Wing Design Automation With ESO and GESO Algis Lencus, Osvaldo M. Querin, Grant P. Steven and Y.M. Xie

Int. J. of Vehicle Design, Vol. 28, No.
3, 2001
Aircraft wing design automation with ESO and GESO
Algis Lencus*, Osvaldo M. Querin
,
,
Grant P. Steven* and Y.M. Xie
)
*
Department of Aeronautical Engineering, University of Sydney,
Bldg. J07, NSW 2006, Australia.
e-mail: algis@aero.usyd.edu.au
e-mail: grant@aero.usyd.edu.au
,
School of Mechanical Engineering, University of Leeds, Leeds LS2
9JT, UK.
e-mail: ozz@mech-eng.leeds.ac.uk
)
School of the Built Environment, Victoria University of
Technology, PO Box 14428, Melbourne City MC, Victoria 8001,
Australia.
e-mail: mike.xie@vu.edu.au
Abstract: Group evolutionary structural optimization (GESO) is a recent
modication of evolutionary structural optimization (ESO) that extends
the topological method to conguration optimization. This paper
demonstrates the optimization of an aircraft wing with ESO and GESO.
The optimization is conducted over some of the major areas involved in the
design of an aircraft wing. These include the optimization of the
conguration of the internal stiffeners, the sizing of the skin thickness
and the detailed optimization of the stiffeners.
Keywords: (To follow)
Reference to this paper should be made as follows: Lencus, A., Querin,
O.M., Steven, G.P. and Xie, Y.M. (2001) Àircraft wing design automation
with ESO and GESO', Int. J. Vehicle Design, Vol. 28, No. 3, pp. 98111.
1 Introduction
Structural optimization methods in general have matured by a large degree and most
can be conveniently applied to many practical problems. Many methods only require
little more than a basic understanding of nite element analysis (FEA) for
implementation of that method. Hence, these methods are aiding the movement
towards `design automation'.
However, there are few examples of the application of a method to the design of a
large scale project. An example of such a project is the design of aircraft wings. These
structures undergo complex loading and the nal design can require months of
development. Aircraft wings are traditionally semi-monocoque structures, usually
comprising an aluminium skin and a large number of folded aluminium stiffeners
oriented in the wing's span-wise and chord-wise directions. Each stiffener is designed
M10012 International Journal of Vehicle Design (IJOVD) Tradespools Ltd., Frome, Somerset
Copyright # 2001 Inderscience Enterprises Ltd.
3B2 Version Number 6.05e/W (Mar 29 1999) {Jobsin}M10056/LIVE 3B2/Article7.3d Date: 31/7/01 Time 09:12am Page 98 of
111
98
with lightness in mind and hence, each is usually made with a number of lightening
holes. It can then be considered that the structural design of an aircraft wing requires
analysis on two major levels, the congurational layout of stiffeners and the detailed
design of each stiffener. If a structural optimization package is to offer an automated
design of an aircraft wing that conforms to traditional aluminium manufacturing
methods, it must work on these two levels. This paper will present how evolutionary
structural optimization (ESO) [1] and a recent modication of ESO, named Group
ESO (GESO) [2], can be used in tandem to fully optimize an aircraft wing.
ESO is a heuristic, ground-based, topology optimization method, and can be
employed as a powerful and robust design tool. It can be used to optimize for a large
number of objectives, including those based on stress, stiffness, frequency and heat.
However, as ESO is inheritantly suited to the optimization of continuum-based
models, like other topology optimization methods it is unsuited to discretized, multi-
component structures. Although an arrangement of components may be ìnter-
preted' from a topology, the results of such an interpretation may often not be
feasible from a manufacturing standpoint [2].
Group ESO was recently conceived as a modication of the ESO method that
allows it to optimize multi-component structures. Group ESO can be used effectively
to constrain the ESO method to only produce manufacturable results. This method
can nd the optimal number of structural members in a model and is capable of
nding the optimal location of those components. In addition, the size of the
structural members, such as the thickness or height can also be optimized [3]. In other
words, GESO offers a means of nding the optimal conguration of components.
The implementation of GESO requires the user to input a set of allowable sizes
and also a `search domain' of possible locations for each structural member. This
search domain of possible locations may be based on what would provide best
`coverage'. In this case, the user would allow for a possible location at evenly spaced
intervals across the design domain in an unbiased fashion. Alternatively, the search
space may be `biased' by allowing for more possible locations in areas that are
expected to be reinforced.
GESO, like ESO, operates on a ground structure basis. Hence, each possible
location for a component is represented in FE by that component. For example, to
represent a search domain that permits the testing of that beam's suitability in four
separate locations, the search domain would be composed of four instances of the
beam in the afore-mentioned locations.
Four variations of ESO were employed in the following example, namely Binary
ESO, Morphing ESO, Binary GESO and Morphing GESO. Descriptions of the
methodology, use and examples of these methods can be found in several references
[2,3,4,5,6]. The design variables that are operated on by Binary ESO are the existence of
elements in an FE model. Binary ESO treats each in a binary fashion, an element may
exist or it may not. Hence, when an element is considered unt by ESO, it is removed.
The design variables used in Morphing ESO are the thickness of elements. Morphing
ESO alters the thickness of unt elements based on a discrete user-dened set of
thicknesses. This method can also remove elements after the smallest thickness is
reached. The two GESO methods are parallels of the two ESO methods, except that the
design variables are not limited to single elements, but user-dened groups of elements.
GESO can then be used to nd the optimal conguration of stiffeners in an
aircraft wing, by working on a ground structure that includes as many stiffeners
111
Aircraft wing design automation with ESO and GESO 99
(which represent possible locations) as can be afforded. However, the method does
not offer detailed design solutions, it is aimed at nding the optimal location and
sizes of pre-dened structural members. Nevertheless, it fulls the requirements of an
important stage of the wing structural design process. A major structural issue that
remains unresolved are uneven stress distributions within structural members and
localized stress concentrations. Although these structural members have their
position and size optimized by GESO, the topology of the component may be too
weak or too strong in some areas, requiring local reinforcing or lightening. Likewise,
areas that experience concentrated loads and `hard points' must also be accounted
for. All these areas require detailed design solutions. For such design problems, a
topological continuum-based method is well suited. ESO can then be used on these
specic areas and stiffeners. In this way, ESO and GESO can be applied in tandem to
offer design solutions for this two-level design problem.
2 Example objective
To exemplify the use of ESO in the congurational stage of a large-scale design
problem, the optimization of a generic twin-engine, propeller-driven light aircraft
wing will be detailed. Specically, the aircraft seats six passengers, is propelled by two
internal combustion engines and includes a retractable landing gear. The maximum
take off weight (MTOW) for this aircraft is approximately 2500 kg, and its wing
surface area is 18.5 m
2
. The design of such an aircraft is to conformto Federal Aviation
Regulations (FAR) 23 regulations [7]. On requirement stipulated in the FAR states
that an aircraft of this type must comply to a `design' load factor of positive 3.8 g. This,
therefore requires that the aircraft must withstand an ùltimate' load factor of 5.7 g.
For this exercise, the wing of the aircraft will be optimized for one load case
that of an ìn ight' condition where the aircraft undergoes a positive 5.7 g load
factor. Added to this is a multiplicative factor of safety of 1.2. This brings the total
loading to positive 6.84 g.
The objective of the optimization is to increase the specic stiffness of the wing for
the afore-mentioned load case. The stiffness can be measured by the inverse of the
structure's mean compliance. The objective of this optimization is then to minimise the
specic compliance, given in (1). To this end, the stiffness sensitivity number (2) [5],
was used as the optimality criteria for all optimization runs. Along with the
maximization of the specic stiffness, the optimization must account for two
constraints. The wing tip must not displace by more than 200 mm, and the wing must
not buckle under its load. These two constraints were not formulated into the
optimization, which is effectively unconstrained. Instead, solutions obtained from the
optimization procedure were compared with these constraints to ensure that they were
not violated. Design variables in the optimization include the number, location and
thickness of the wing's internal stiffening members and the thickness of the skin.
Specific compliance = C * W (1)
where C is compliance and W is weight.
a
i
=
1
2
u
i

T
K
i

u
i

(2)
111
100 A. Lencus, O.M. Querin, G.P. Steven and Y.M. Xie
where a
i
is the stiffness sensitivity number for element i, u
i
is the displacement
vector for element i and K
i
[ [ is the stiffness matrix for element i.
3 Modelling details
A nite element model of the aircraft is shown in Figure 1. As can be seen, the
symmetry of the aircraft was used to the advantage of the FE model. In addition,
only the part of the fuselage in the vicinity of the wing was modelled. To
approximate the ight condition, the fuselage was attached to a number of beams,
which in turn, were attached to fully xed points to effectively suspend the model
and prevent free translation.
At a load factor of 6.84 g, the wings must provide lift for the entire weight of the
aircraft, which at MTOW is 167580 N. With a wing surface area of 18.5 m
2
, the
average lifting pressure on the wing is approximately 9 kPa. This aerodynamic load
was modelled chord-wise with a triangular distribution on both the upper and lower
surface of the wing, and a separate leading edge pressure as shown in Figure 2a. The
span-wise distribution was modelled with an elliptical distribution with extra loading
around the propeller area to account for accelerated ow, Figure 2b.
111
Figure 1 FE model of the twin-engine light aircraft.
Figure 2 Chord-wise and span-wise wing pressure distribution.
Each engine and propeller combination weighs 185 kg, at 6.84 g each applies a
force of 12.4 kN on the wing's engine mount. It was assumed that the thrust due to
the propellers in this load case was 4 kN. For modelling purposes, it was assumed
that all parts of the aircraft were constructed out of sheet aluminium with a density
of 3.2 Mg/m
3
. This means that the density is a constant in (1), which lets the specic
compliance calculated for this example to be based on (3), where volume is
substituted for weight. The aps and ailerons of the aircraft were not modelled, as
they were not intended to take part in the optimization.
Specific compliance = C *V (3)
where C is compliance and V is Volume.
Care was taken to ensure that the model had a minimum number of degrees of
freedom and a minimum bandwidth to expedite optimization without affecting the
quality of the wing analysis. The simplicity of the model must be taken in context
that this is a demonstrational model of a hypothetical aircraft compromised only by
the amount of time available to detail the model and the time and hardware
requirements to otherwise solve a more elaborate model.
4 Optimization procedure
The stiffness sensitivity number of (2) was used as the optimality criteria. Instead of
using the Rejection Ratio [4] commonly used with ESO, a sorting technique was
employed to obtain the correct deletion criteria. This method sorts the optimality
criteria values for all the design variables into a series sorted in descending order.
This then provides for an easily controllable rate of removal per iteration. If 57 of
design variables (with the smallest sensitivity values) are to be removed per iteration,
the sensitivity value of the variable that is at the bottom 57 position of the sorted
series is used as the deletion criteria. All design variables with an optimality criterion
less than the deletion criterion are then removed.
Optimization of the aircraft wing consisted of the following steps:
1 Application of Binary GESO to obtain locations of internal stiffeners.
2 Sizing optimization of stiffeners with Morphing GESO to obtain thicknesses.
3 Sizing optimization of skin with Morphing ESO to obtain thicknesses.
4 Detailed optimization of a stiffener with Binary ESO.
Step 4 involved the use of a separate FE model with a nely meshed stiffener for
detailed topology optimization. Because the optimization for this stage was directed
at only one stiffener, including the rest of the aircraft's structure would be
unnecessary, and computationally expensive. Instead, a sub-model of the volume of
interest was created. This was achieved by obtaining from the global model, the
solved displacements at the boundary of the region to be sub-modelled. Then
prescribing these displacements as constraints on the sub-model's boundaries. This
allowed for detailed analysis of the section without the computational overhead of
including the remaining parts of the global model. The creation of this sub-model
was performed in Strand 7 [8].
111
A history of the mean compliance, volume and wing tip displacement of the
aircraft model was recorded for all iterations. It must be noted that the applied
pressure on the wing remained unchanged through the optimization even though the
weight of the wing was continually reduced.
To ensure that the buckling constraint was not violated, a buckling FEA was
performed on selected solutions obtained from the optimization runs. A buckling
analysis was not performed at every iteration as the cost of a buckling solution was
considered too great. For this example, if buckling occurs in highly localized regions of
a stiffness-optimized model, the region was manually reinforced to prevent buckling
by assigning a greater thickness to the skin in that region. However, if buckling occurs
across a signicant area, that model was not considered for manual reinforcement,
and hence, ignored. It must be noted that in this example, the ground structure mesh
shown in Figure 3 does not accommodate for local buckling analysis, as there are no
nodes in between stiffener intersections to allow for out-of-plane movement. This
situation however, changes as stiffeners are removed. Regardless, a correct local
buckling analysis can only be performed with a more detailed model. Such an analysis
however, is beyond the scope of this example of a preliminary design stage.
5 Ground structure for Group ESO
A ground structure of the aircraft was prepared for Group ESO with a large number
of internal web stiffeners in the wing, Figure 3. These stiffeners represent the fully
populated design domain for Group ESO. A total of 137 separate lengths of stiffeners
were included in this domain. Note that there are two areas near the wing root that are
devoid of stiffeners; this is to allow a recess for the retractable undercarriage.
Each stiffener length was assigned a separate group number. All other areas
including the fuselage and wing skin were assigned a non-design group number to
ensure that these areas remained unchanged. The span-wise members were split into
a number of semi-lengths measuring roughly one quarter of the span of the half-
wing. Otherwise, all chord-wise and diagonal members run uninterrupted from
leading to trailing edge. The thickness of the skin (except in critical areas near the
wing root) was set to 3.5 mm. All stiffeners were given a thickness of 3 mm. Although
most areas of the wing would not require the use of such thick members, other areas
111
Figure 3 Ground structure for Group ESO.
closer to the wing root experience greater loads. Nevertheless, the use of (on average)
overly thick members is in keeping with a top-down ground structure based
method.
The ground structure is not intended to be a possible design, this would be
implausibly heavy with a stiffener thickness of 3.0 mm. Furthermore, a wing for this
type of aircraft should allow a large cavity for an internal fuel tank. Instead, GESO
employed this ground structure as a basis from which to obtain an arrangement of a
relatively small number of stiffeners. In all likelihood, this nal arrangement would
then provide large enough cavities for a fuel tank.
Figure 3 shows that the wing tip included a relatively small number of stiffeners
as it was of less concern; the GESO search technique was not `directed' at this area.
The use of diagonal stiffeners in wings is unusual in aircraft of this type. Diagonals
were employed to increase the number of locations available to the GESO `search
technique' whilst not increasing the model's degrees of freedom and thereby having a
minimal impact on the analysis bandwidth. The use of diagonals also provided the
opportunity to see whether they are preferable to traditional longitudinal members.
6 Results
Initially, Binary GESO was applied to the model such that 17 of groups were
removed per iteration. This usually resulted in the removal of one stiffener per
iteration. The compliance and volume values for the model were recorded
throughout the iteration history. Figure 4 shows the value of specic compliance
(3) for this optimization run. From this graph, it can be seen that the specic
compliance reaches a minimum at iteration 70. Figure 5 shows the stiffener
conguration at this iteration.
To further optimize the conguration, Morphing GESO was applied to the
model to size optimize the thickness of the stiffeners such that 17 of groups were
sized per iteration. The discrete set of thicknesses from which to size the stiffeners is
given in Table 1. The structure given in Figure 3 was not used as the ground
structure for this run, instead Morphing GESO was used in a sequential fashion.
Nevertheless, the structure in Figure 5 was not used either as it may not contain
features that would be kept had thinner stiffeners been available in the Binary GESO
run. Instead, a structure with a smaller volume reduction, from iteration 56 (shown
in Figure 6) , was employed as the ground structure for the Morphing GESO run.
The specic compliance results for this optimization run are shown in Figure 7.
Iteration 105 offers the minimum specic compliance and the stiffener conguration
it suggests is shown in Figure 8. Note that all but one stiffener is of 3.0 mm thickness.
With the conguration of the internal stiffeners optimized, Morphing ESO was
then applied to size the thickness of the skin to further optimize the wing model. A
111
Table 1 Set of stiffener thicknesses used by Morphing GESO.
Property 1 2 3 4 5 6 7 8 9 10 11
Thickness (mm) 3.0 2.54 2.03 1.8 1.57 1.27 1.02 0.81 0.64 0.51 0.41
number of critical areas on the skin (near the wing root) were assigned as part of the
non-design domain. Due to the larger number of variables; in this case every element
that represented the skin; Morphing ESO was set to size 57of elements per iteration.
The discrete set of skin thicknesses available for this optimization run are shown in
Table 2. The ground structure chosen for this procedure was that of Iteration 63,
111
Figure 5 Result of Binary GESO at iteration 70.
Figure 6 Result of Binary GESO at iteration 56.
Figure 4 Specic compliance history for the Binary GESO run.
Figure 9. The structure of Figure 8 was not used as it violated the buckling
constraint.
The specic compliance history for this run is shown in Figure 10. The optimum
was reached at iteration 301. Figures 11 and 12 show the post-processed skin
thickness conguration for the top and bottom surface of the wing at this iteration,
respectively.
With the conguration of the wing optimized, the design was then ready for
detailed optimization of the stiffeners. For this example, only one stiffener was
optimized. A sub-model of a small section of the middle region of the wing was
created, as shown in Figure 13. This sub-model includes part of the skin, and a
number of stiffeners. Only one of the stiffeners is wholly contained in this model; it is
this stiffener that was optimized.
111
Figure 7 Specic compliance history of Morphing GESO run.
Figure 8 Result of Morphing GESO at iteration 105.
Table 2 Set of skin thicknesses used by Morphing ESO.
Property 1 2 3 4 5 6 7 8
Thickness (mm) 3.5 3.0 2.54 2.03 1.8 1.57 1.27 1.02
The sub-model includes the region that surrounds the stiffener in question to
ensure that any changes in the stiffener over the course of the optimization will have
a minimal effect on true displacements at the constrained nodes. This is important
because the constraints remain unchanged during the evolution. The determination
of how much volume to include in the sub-model was then a compromise between
111
Figure 9 Result of Morphing GESO at iteration 63.
Figure 10 Specic compliance history of Morphing ESO run.
Figure 11 Upper skin thickness conguration at iteration 301.
the computational cost of additional elements and the quality of the constraints over
the course of the optimization.
The stiffener to be optimized was re-meshed to include a total of 1152 shell plate
elements. ESO was then applied to only this stiffener to obtain an optimal topology.
ESO was set to treat elements in a binary fashion, and to remove 0.57 of elements
per iteration based on stiffness sensitivity. The normalized specic compliance
history for this evolution is shown in Figure 14. Iteration 114 offers the minimum
specic compliance; the topology for this iteration is shown in Figure 15. A more
detailed picture of the result, and a spline tted outline of this topology is given in
Figure 16.
7 Discussion
The optimization of the conguration of the stiffeners, the skin thickness and the
detailed optimization of the stiffener all resulted in a signicant reduction of the
specic compliance, and therefore, the increase in specic stiffness. The total increase
in specic stiffness of the structural model shown in Figures 11 and 12 is
approximately 407, whereas the total volume was reduced by 327. Initially,
the ground structure of Figure 3 had a maximum wing tip displacement of 7.7 mm.
111
Figure 12 Lower skin thickness conguration at iteration 301.
Figure 13 Sub-model ground structure used by Binary ESO.
The structure of Figures 11 and 12 had a maximum of 97.8 mm, which is well
within the 200 mm constraint. This model also resulted in a buckling eigenvalue of
1.1 under the ultimate load condition.
All stiffener congurations suggested by the two GESO runs showed a strong
preference for diagonal stiffeners over span-wise spars and chord-wise ribs. This
111
Figure 14 Normalized specic compliance history for Binary ESO run.
Figure 15 Result of Binary ESO at iteration 114.
Figure 16 Spline-tted stiffener topology at iteration 114.
preference for diagonal stiffeners was also evidenced in the examples given in early
studies of Group ESO applied to similar structures [3]. It can be suggested that this
occurs as the diagonal members assumed the role of both types of orthogonal
stiffeners. The diagonals maintain the aerofoil cross sectional shape, reduce
compressive lengths and distribute loading much like a rib would. They also carry
bending loads as a spar would. Nevertheless, if manufacturing methods do not
permit the use of diagonal members, the ground structure can be comprised of only
orthogonal members to support this constraint.
Application of detailed optimization to the stiffener resulted in a stiffener that
was shortened such that it ended off at intersections with other stiffeners. The
stiffener chosen was a span-wise member that would carry predominantly bending
loads and which provides some interface for roughly uniform loads between the top
and bottom skins. Due to these factors, the resultant topology is much like two
stringers attached to the top and bottom skins with a thin webbing connecting the
two. It was assumed that the chequerboard region represents thinner material. The
spline-tted topology shows where lightening holes may be placed and the broken-
lined spline where the thinner webbing may be used.
Detailed optimization of the aircraft wing does not need to be limited to the
stiffeners. Morphing ESO can be applied the local areas of the skin to determine a
thickness distribution that can be interpreted as the placement of stringers.
Further improvement of these results can be attained through a number of
means. The ground structure used by GESO in the example was arbitrarily chosen
with stiffeners placed at regular intervals. Prior knowledge of an improved design
can be used to create a more rened ground structure that may offer better results.
This could be achieved as a second stage after obtaining the results offered in the
Binary GESO run in this example, thus `targeting' the search towards an optimum.
Alternatively, a ground structure with a larger number of stiffeners would offer
superior results in the same way a ne mesh would for ESO as it effectively allows for
the testing of components in a greater number of locations.
8 Conclusion
The example in this paper demonstrates the ease of the application of the ESO based
methods. Besides the ability to produce an accurate FEA model that correctly
represents the applied loading a structure is to be designed for, ESO requires very
little else from the user to obtain good results. The example shows that this is true
even when the optimization problem is a large scale one of two levels of complexity,
such as an aircraft wing. This conrms that the ESO and GESO methods are viable
design automation tools.
References
1 Xie, Y.M. and Steven, G.P. (1997) Èvolutionary Structural Optimization', Springer-
Verlag, Berlin.
2 Lencus, A., Querin, O.M., Steven, G.P. and Xie, Y.M. (1999) `Modications to the
Evolutionary Structural Optimization (ESO) Method to Support Congurational
111
Optimization', 06-EVM1-5, World Congress of Structural and Multidisciplinary
Optimization, May 1721, Niagara Falls, New York.
3 Lencus, A., Querin, O.M., Steven, G.P. and Xie, Y.M. (1990) `Group ESO with
Morphing', 1st ASMO UK/ISSMO Conference on Engineering Design Optimization, July
89, Ilkley, West Yorkshire.
4 Steven, G.P. and Xie Y.M. (1993) À simple evolutionary procedure for structural
optimization', Computers & Structures, Vol. 49, No. 5, pp. 885896.
5 Nha, C.D., Xie, Y.M. and Steven, G.P. (1998) Àn evolutionary structural optimization
method for sizing problems with discrete design variables', Computers & Structures, Vol.
68, No. 4, pp. 419431.
6 Querin, O.M. (1997) Èvolutionary Structural Optimization: Stress Based Formulation
and Implementation', PhD thesis, Department of Aeronautical Engineering, University of
Sydney, Australia.
7 Federal Aviation Regulations, Part 23, Department of Transportation, Federal Aviation
Administration, Distribution Requirements Section M-482.2, Washington D.C., USA.
8 GD Computing Pty. Ltd., Sydney, Australia.
111

Aircraft Wing Design Automation With ESO and GESO Algis Lencus, Osvaldo M. Querin, Grant P. Steven and Y.M. Xie

Enviado por

Dados do documento

Descrição original:

Título original

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Aircraft Wing Design Automation With ESO and GESO Algis Lencus, Osvaldo M. Querin, Grant P. Steven and Y.M. Xie

Enviado por

Direitos autorais:

Formatos disponíveis

Int. J. of Vehicle Design, Vol. 28, No.

Você também pode gostar