A R C H I V E O F M E C H A N I C A L E N G I N E E R I N G

VOL. LXII 2015 Number 1

10.1515/meceng-2015-0003
Key words: box wing, joined wing, optimization, conceptual design

MIŁOSZ J. KALINOWSKI ∗

STRUCTURAL OPTIMIZATION OF BOX WING AIRCRAFT

The box wing system is an unconventional way to connect the lifting surfaces
that the designers willingly to use in prototypes of new aircrafts. The article present
a way to quickly optimize the wing structure of box wing airplane that can be
useful during conceptual design. At the beginning, there is presented theory and
methods used to code optimization program. Structure analysis is based on FEM
beam model, which is sufficient in conceptual design. Optimization is performed
using hybrid method, connection of simple iteration and gradient descent methods.
Finally, the program is validated by case study.

Notation
Cx – drag coefficient
Cm – moment coefficient
Cz – lift coefficient
ρ – air density
n – load factor
I x , Iy , I xy – section moments of inertia
q – dynamic pressure
V – aircraft speed
σallow – allowable stress

1. Introduction

The general concept of box wing configuration was presented by Prandtl

in 1924 [1]. Many potential advantages as well as innovation of the design
are the reasons why this configuration has become more and more popular.
∗
Institute of Aviation, Warsaw University of Technology, Warsaw, Poland; E-mail:
[email protected]

- 10.1515/meceng-2015-0003
Downloaded from PubFactory at 08/22/2016 11:28:00PM
via free access
46 MIŁOSZ J. KALINOWSKI

Theoretically, it should allow for reducing of both the empty weight and
aerodynamic drag [2]. Both effects can reduce the fuel consumption, thus
decreasing the operational costs and make aircraft more ecological. But, one
should take into account a number of problems that are confronted with when
calculating and choosing dimensions of such a structure, before deciding to
use that solution. Static indetermination and non-linearity can seriously com-
plicate the task. An additional difficulty consists in the design optimization
taking into account the imposed requirements. Given these problems, it be-
comes practically impossible to carry out the task of choosing dimensions
of the real structure with the use of analytical calculations only. Fortunately,
modern technology supports us yielding various programming capabilities.
Research of Gallman, Smith and Kroo [2] could serve as an example of
automated optimization of similar joined wing configuration. They used sev-
eral of commercial programs to perform the task. The optimization method
adopted by them was based on flow calculations using the LinAir software
package. The NPSOL program was responsible for the optimization process.
The interface created by Gallman at all. controlled the calculation process.
The flow around the wing was calculated using the vortex-lattice method
(LinAir). As a result, one could get the aerodynamic forces acting on the
wing. These were used as an input to another software. Gallman at all.
considered multiple load cases and performed the calculations of basic flight
mechanics. That allowed for optimization of 17 design variables, including
mechanical features, as constraints.
A different approach to solving optimization problems is presented in [3].
The authors decided to include phenomena such as aeroelastic failures and
instability, in calculations relevant to statically indeterminate configuration.
Ensuring from the very beginning of project that such a phenomena will not
appear, is the key to success.
Linear approach to the problem of highly loaded surface leads to signif-
icant errors. There is therefore a need to develop a method of the structure
optimization taking structural nonlinearity into account. The NROEL method
[4, 5, 6] could be an example. This is a modified gradient method, which
transforms the nonlinear problem to the linear one and then solves it. This
method consists in calculating the equivalent loads for the linear approach
that causes the same displacements as in nonlinear models.

2. General Survey

The mission is the first assumption which is made during the design of
new aircraft, which in turn defines the required parameters of the airplane.
In most cases, the main requirement consist in the lowest possible empty

- 10.1515/meceng-2015-0003
Downloaded from PubFactory at 08/22/2016 11:28:00PM
via free access
 STRUCTURAL OPTIMIZATION OF BOX WING AIRCRAFT 47

weight and the largest payload with the specified size. The loads acting on
the structure are defined mainly by its size. Larger forces require a stronger
structure which may cause an increase in mass. As a consequence, the design
process becomes iterative. There is therefore a need to reduce the structure
mass to a minimum. Such an action provides convergence to the solution
and reduces costs.
The POSUPS is an example of a software code that solves this task. The
code was written by the author in Delphi and its interface allows for co-
operation with commercial software. A three-dimensional linear FEM model
was created, in which the wings are represented by beams of variable cross
sections. The wing cross-section was modeled as one-cell wing box with four
flanges placed in the corners. The opposite walls have the same thickness, and
the cross-sections of the flanges located on the opposite corners are identical
(Fig. 3). This configuration of structure is considered in the literature as
optimal for the joined wing [7]. For this reason the author decided to use it
in POSUPS program.

Fig. 1. Block diagram of POSUPS program

- 10.1515/meceng-2015-0003
Downloaded from PubFactory at 08/22/2016 11:28:00PM
via free access
48 MIŁOSZ J. KALINOWSKI

The aerodynamic loads can be obtained from any program that calculates
the aerodynamic coefficients distributions along on the wing span. When
the loads are introduced from the file, they are approximated for the FEM
model using the Legendre polynomials. The final static analysis solution is
obtained using the Gaussian elimination method. This method provides a
reliable solution.
Optimization of the structure is performed by taking the value of actual
mass as the objective function. On the other hand, wall thickness and flanges
cross-section area serve as optimization variables. Searching for the optimal
structure is carried out for the critical load. To limit the area of analysis, the
penalty function is applied. It is a quadratic form that uses stress to determine
the value of penalty. The algorithm which searches for a minimum weight is
based on two methods. The first one is the gradient descent method, being
one of gradient methods. The second one is a simple iteration method based
on the fixed-point theorem. A combination of both methods in order to take
advantage of both algorithms at once is the best way to optimize the structure
of the aircraft. Distributions of cross sections, forces, displacements and
stresses corresponding to the minimum weight of the structure are obtained
as a result of searching for the optimal solution. Due to the simplification
and numerous assumptions accepted, the presented software code is suitable
for the analysis of small unmanned aircraft, or a conceptual analysis of larger
structures.

3. Wing Model

The model used in conceptual design should be the simplest one, in

which the number of variables is small and most intuitive. As a result, quick,
rough estimation of the structure is possible. In the next calculation stages a
simple model is subject to comprehensive local analysis. Therefore, the box
wing configuration (Fig. 2) is represented in terms of a spatial arrangement
of beams [3, 7]. Each beam represents a single lifting surface: two front
surfaces, two rear and tip plates, respectively. The front surfaces are arranged
in a high wing configuration, since in the literature some suggestions about
additional aerodynamic advantages of that configuration [8] can be found.
Each beam is fixed at the quarter of chord from its leading edge. This three-
dimensional beam model adequately represents the design for the conceptual
design purposes.
To maintain the problem simplicity, it is assumed that the internal struc-
ture of the wing can be represented by a one-cell torsion box only. The
rationale for this approach consists in the fact that the stiffness of other
elements of the wing (including wing skins) is usually much lower than the

- 10.1515/meceng-2015-0003
Downloaded from PubFactory at 08/22/2016 11:28:00PM
via free access
 STRUCTURAL OPTIMIZATION OF BOX WING AIRCRAFT 49

Fig. 2. Geometry of the analyzed box wing

stiffness of structural elements. It means that the loads carried by them are
small as compared to the forces acting upon the structural elements. The
torsion box (Fig. 3) consists of two horizontal and two vertical webs, and
four flanges that are placed in the corners. It was assumed that the opposite
flanges have the same cross-sectional area, which is due to the fact that the
resultant moment acting on the cross-section lies in the plane of the section
but does not coincide with the axes of local reference system. The objective is
therefore to ensure that the resultant moment acts along one of the principal
axes of inertia.

Fig. 3. Geometry of the wing box section

Assuming structure model illustrated above, one can determine the ap-
proximate values of the moments of inertia for each of the current wing
section. These values are determined from the formulas [3]:

- 10.1515/meceng-2015-0003
Downloaded from PubFactory at 08/22/2016 11:28:00PM
via free access
50 MIŁOSZ J. KALINOWSKI

2h2 l2 δ1 (ξ) δ2 (ξ)

I x (ξ) = (1)
lδ2(ξ) + hδ1 (ξ)

!2
l
Iy (ξ) = 2 (A1 (ξ) + A2 (ξ)) (2)
2

!2
h
Iz (ξ) = 2 (A1 (ξ) + A2 (ξ)) (3)
2

hl
Iyz (ξ) = (A1 (ξ) − A2 (ξ)) (4)
2
Note that the cross section is asymmetric. As a result, directions of the
principal axes of inertia should be determined. Moments of inertia should be
obtained with respect to these axes. The best way to do this is to use Mohr’s
circle for the moments of inertia.

4. Loads Mapping Onto FEM Model

The analysis is performed for the predefined load distributions, which

result from the distributions of aerodynamic coefficients along each lifting
surface. They can be obtained by using the aerodynamic software for calcu-
lation of aerodynamic loads for specified flight conditions (e.g. AVL). The
distributions obtained from an available aerodynamic software are usually
defined for a mesh different from that used in the FEM software. For that
reason, it is necessary to transform the distributions of coefficients resulting
from the aerodynamic software to the mesh applied by the POSUPS.
The transformation of loads is performed using the 8th and 9th degree
Legendre polynomial approximation [9]. The following recursive formula is
used to obtain the polynomials:

P0 (x) = 1

P1 (x) = x (5)
2n + 1 n
Pn+1 (x) = xPn (x) − xPn−1 (x)
n+1 n+1

- 10.1515/meceng-2015-0003
Downloaded from PubFactory at 08/22/2016 11:28:00PM
via free access
 STRUCTURAL OPTIMIZATION OF BOX WING AIRCRAFT 51

The main advantage of this method consists in its correct performance

even in the case when the matrix used to determine multipliers of polynomial
is ill-conditioned.

5. FEM Formulation

The algorithm that solves the problem of static analysis employs the
finite element method based on linear displacement equation (6). Currently,
in most available codes this approach is used in analysis of the structure. The
main advantages of this algorithm are simplicity of implementation into any
programming language and the possibility of calculation of complex shapes.

[K] {Q} = {F} (6)

The FEM model consists of a frame formed from beam elements. Each
element has a characteristic cross section defined in Fig. 3. For these cross
sections the stiffness matrix [K] is determined.
The spatial arrangement of beams is loaded by fluxes of aerodynamic
forces:
fz = qCz (x) c (x)
f x = qC x (x) c (x) (7)
fm = qCm (x) c2 (x)

where q = 0.5ρV 2 represents the dynamic pressure.

Since a continuous load distributed along each element is considered,
there is a need to define the corresponding nodal forces according to the
relation [10]:
Z le
Pi = f j (ξ) Nk (ξ) dξ (8)
0

Applying the loads to the nodes, one should pay attention to vector directions.
The box wing is a symmetrical system. It allows for a half-oft-the-wing
analysis only, assuming the necessary boundary conditions. The above sim-
plification reduces significantly the computation time because there is no
need to analyze the entire model.
The Gauss elimination method is used to solve displacement equation
(6). This method is used because of its simplicity and versatility. In contrast
to the iterative methods, we always obtain the solution of the system of
equations (provided that it exists). There is no problem with the selection of
initial values or convergence of the solution.

- 10.1515/meceng-2015-0003
Downloaded from PubFactory at 08/22/2016 11:28:00PM
via free access
52 MIŁOSZ J. KALINOWSKI

Fig. 4. Description of the current section

One can calculate the internal forces acting upon sections based on the
displacements obtained using the Gaussian elimination. Using these forces,
one can determine the stresses in flanges and webs in every cross section of
the torsion box.
Normal stresses in flanges:
!
h l N
(1) : σ = σM − , +
2 2 2 (A1 + A2 )
!
h l N
(2) : σ = σM − , − +
2 2 (A
2 1 + A2 )
! (9)
h l N
(3) : σ = σM , − +
2 2 2 (A1 + A2 )
!
h l N
(4) : σ = σM , +
2 2 2 (A1 + A2 )
where [11]:
Mz Iyz + My Iz Mz Iy + My Iyz
σM (y, z) = 2
z − 2
y (10)
Iz Iy − Iyz Iz Iy − Iyz
Shear stresses in webs:
!
1 Ms
(I) : τ = Tz −
2δ1 l h
!
1 Ms
(III) : τ = Tz +
2δ1 l h
! (11)
1 Ms
(II) : τ = Ty −
2δ2 h l
!
1 Ms
(IV) : τ = Ty −
2δ2 h l

- 10.1515/meceng-2015-0003
Downloaded from PubFactory at 08/22/2016 11:28:00PM
via free access
 STRUCTURAL OPTIMIZATION OF BOX WING AIRCRAFT 53

6. Optimization Algorithms

One should use various types of tricks to cope effectively with the task
consisting in finding a minimum of a function of several variables. Most
commonly those tricks are used in the problem definition (e.g. solving an
equivalent problem) and then the approach is applicable only to the given
specific case.
To simplify the current task, one can replace the weight minimization
with an equivalent problem. Searching for optimum cross sections of a pre-
defined structure is equivalent to the problem of searching for the maximum
material effort. At first glance, this is a very similar task. However, in this
case cross sections are selected to achieve the predetermined allowable stress
in each section.
In the present optimization problem two methods are used: simple iter-
ation method [12] and gradient descent method [13]. The simple iteration
method is used to estimate the optimum quickly but not accurately, while the
gradient descent method allows for reaching the final solution. The simple
iteration algorithm is based on the Banach theorem of fixed point. This is
a very fast convergent algorithm and its computational cost is proportional
to the number of variables. Another but not less important advantage of this
method consists in its flexibility in choosing the starting point. The conver-
gence is always achieved, no matter from which point the calculation will
start. This is the convergence to the same solution.

Fig. 5. Simple iteration algorithm

Searching for the solution is performed following the formula:

σn
An+1 = An (12)
σdop

- 10.1515/meceng-2015-0003
Downloaded from PubFactory at 08/22/2016 11:28:00PM
via free access
54 MIŁOSZ J. KALINOWSKI

From relation (12) it can be clearly seen that the optimal solution can
be reached just after the first step of approximation for statically determinate
systems,. Unfortunately, the considered box wing configuration is statically
indeterminate. In reality, changes in cross-sections are not linear as those
obtained from equation (12). Stresses in the current cross section depend on
the stiffness of the wing in each section. But when one considers this problem
step by step according to (12) one can expect convergence to minimum values
of the variables in each section. However, there is one more obstacle. While
for the statically determined case the assumptions of Banach theorem are
always true, checking if these assumptions are fulfilled for complex and
statically indeterminate cases is practically impossible. Nevertheless, based
on experience gained in solving similar problems, one can assume that this
algorithm does not tend to diverge. The worst case happens when the solution
oscillates around the final solution. Then, a different method has to be used
to carry out several successive iterations.
The best method to complete the iteration is the simplest one. In this
case the gradient descent method was chosen. This algorithm has the lowest
computational cost among all gradient methods, but its convergence can be
very slow. The solution is searched in the opposite direction to that indicated
by the objective function gradient.
The gradient descent algorithm is much less effective than a simple
iteration method because one has to determine not only the value of the
objective function but also its derivatives at each step. However, this method
suits perfectly for finishing the iterative solution. The objective function f (x)
given above provides constraints due to the maximum allowable stress in the
cross section for the gradient descent method.

f (x) = m (x) + c · S (x) (13)

Objective function (13) is the sum of structure mass and penalty function.
The latter is a quadratic form of stresses appearing in the cross section. Its
value is determined from the formula:
 N


 X 2


 σ i (x) − σdop if σ i (x) > σallow
S (x) = 
 (14)



i=1
 0 otherwise

Solution to the problem consists in finding an approximate solution (us-

ing simple iteration algorithm) and then obtaining more accurate solution
with the use of gradient descent method.

- 10.1515/meceng-2015-0003
Downloaded from PubFactory at 08/22/2016 11:28:00PM
via free access
 STRUCTURAL OPTIMIZATION OF BOX WING AIRCRAFT 55

Fig. 6. Gradient descent algorithm

7. Case Study

The Author decided to optimize the structure of an unmanned aircraft

wing shown in Fig. 7 to verify working correctness of the POSUPS software.

Fig. 7. Geometry of analyzed UAV

- 10.1515/meceng-2015-0003
Downloaded from PubFactory at 08/22/2016 11:28:00PM
via free access
56 MIŁOSZ J. KALINOWSKI

NACA 23012 airfoil is used in this aircraft. The analysis was carried out for
critical flight conditions (maximum lift and maximum load factor):
• n = 6.58
• V = 44.5 m/s
In the calculations it was assumed that the length of torsion box section
is 35% of chord, while the height is 11% of chord. To simplify the problem,
aluminum alloy 2024 was selected for the wing structure. The only constraint
for optimization is an allowable stress of the material.
The AVL software was used to determine the loads acting upon the wing.
The following distributions of aerodynamic coefficients [14] were obtained.

Fig. 8. Distribution of Cz coefficient along the wing span

Fig. 9. Distribution of Cx coefficient along the wing span

- 10.1515/meceng-2015-0003
Downloaded from PubFactory at 08/22/2016 11:28:00PM
via free access
 STRUCTURAL OPTIMIZATION OF BOX WING AIRCRAFT 57

Fig. 10. Distribution of Cm coefficient along the wing span

The FEM model developed for optimization purposes consisted of 30

finite elements, 10 on each lifting surface. This effectively gives 120 de-
sign variables (spar caps cross section area and webs thickness). Iterative
calculations of the structure dimensions started from the same fixed initial
values in each section. The wall thickness of 0.5 mm was applied, the initial
cross section area of the flanges was 40 mm2 each. The starting values were
chosen based on stress values in each element. It was assumed that half
of them should give stress values greater than the allowable stress. There
were 450 iterations of POSUPS program performed. First 50 iterations were
performed using the simple iteration, while the descent gradient method was
employed in the rest part of them. The results of the program are shown in
Fig. 11 and Fig. 12.

Fig. 11. Distribution of stress in all elements

- 10.1515/meceng-2015-0003
Downloaded from PubFactory at 08/22/2016 11:28:00PM
via free access
58 MIŁOSZ J. KALINOWSKI

As it can be seen from Fig. 12 simple iteration method quickly leads

to the solution area and then oscillates around it. Further calculations using
gradient method lead to convergence solutions. Fig. 11 shows that the optimal
solution has not been finally achieved. Moreover, there are areas where the
stress exceeds the allowable values. This error does not exceed 5%. To get
rid of these areas the optimization should be continued.

Fig. 12. Changes of mass and objective function

During the optimization it turned out that the stiffness of regions con-
necting the lifting surfaces is practically the most important in terms of a
deformation of the entire box wing system. The less stiff joining region the
more it resembles articulated connection. It causes high sensitivity in the
whole structure to any changes in this region. Even a small change in this
region causes rapid changes of forces and stresses in the whole structure and
poses problems with convergence.

8. Conclusions

The software code based on linear formulas has been developed to opti-
mize the structure of box wing aircraft. The code employs the aerodynamic
loads obtained using the aerodynamic software available on the internet. The
approximation method based on the Legendre polynomials gives good results
for approximation of aerodynamic coefficient distributions, therefore it is well
suited for mapping loads from external software onto the FEM model created
in the code. The strength calculations were performed using the FEM based
on the 3D beam model. The Gauss elimination method was used to obtain
the solution. Although it is rather a time-consuming algorithm, there are no
such problems with convergence as in the case of iterative algorithms.
The optimization process was performed with the application of two
methods: simple iteration and gradient descent. While these methods solve

- 10.1515/meceng-2015-0003
Downloaded from PubFactory at 08/22/2016 11:28:00PM
via free access
 STRUCTURAL OPTIMIZATION OF BOX WING AIRCRAFT 59

statically determinate problems very well, statically indeterminate analysis

can pose serious obstacles for both of them. The hybrid method which is a
connection of both methods combines good qualities of both algorithms and
can cope with the problem of optimizing the structure of box wing aircraft.
The structure optimization of UAV wing gave reasonable results, paving
a good way for the concept of an aircraft wing structure in the box wing
system. The analysis also showed a disturbing tendency of the solution in
the areas of lifting surfaces connection.

Manuscript received by Editorial Board, September 17, 2013;

final version, February 02, 2015.

REFERENCES

[1] Prandtl L.: Induced drag of multiplanes. NACA Report no. 182, USA, March 1924.
[2] Gallman J.W., Smith S.C., Kroo I.M.: Optimization of Joined-Wing Aircraft. Journal of
Aircraft, 1993, Vol. 30, No. 6, pp. 897-905.
[3] Bindolino G., Ghiringhelli G., Ricci S., et al.: Multilevel Structural Optimization for Prelim-
inary Wing-Box Weight Estimation. Journal of Aircraft, 2010, Vol. 47, No. 2, pp. 475-489.
[4] Park G.J.: Nonlinear Response Structural Optimization of a Joined-Wing Using Equivalent
Loads. Report, Hanyang University, Korea, 15 November 2007.
[5] Kim Y.I., Park G.J., Kolonay R.M., et al.: Nonlinear Response Structural Optimization of a
Joined-Wing Using Equivalent Loads. AIAA Journal, 2008, Vol. 46, No. 11, pp. 2703-2713.
[6] Green N.S.: Structural Optimization of Joined-Wing Beam Model with Bend-Twist Coupling
Using Equivalent Static Loads. Master Thesis, Air Force Institute of Technology, USA, 2009.
[7] Wolkovitch J.: The Joined Wing. An Overview. Journal of Aircraft, 1986, Vol. 23, No. 3,
pp. 161-178.
[8] Mamla P., Galiński C.: Basic induced drag study of the joined-wing aircraft. Journal of
Aircraft, 2009; Vol. 46, No. 4, pp. 1438-1440.
[9] Baron B., Pasierbek A., Maciążek M.: Algorytmy numeryczne w Delphi. Księga eksperta.
Gliwice, Helion, 2006.
[10] Liu G.R., Quek S.S.: The finite element method. A practical course. Butterworth-Heinemann,
2003.
[11] Niu M.C.Y.: Airframe stress analysis and sizing.AD Adaso/Adastra Engineering LLC, Hong
Kong, 1999.
[12] Granas A., Dugundji J.: Fixed Point Theory. New York, Springer, 2003.
[13] Kusiak J., Danielewska-Tułecka A.: Optymalizacja. Wybrane metody z przykładami zas-
tosowań, Warszawa, PWN, 2009.
[14] Mach A.: Projekt koncepcyjny samolotu bezzałogowego do patrolowania lasów. Master The-
sis, Warsaw University of Technology, Poland, 2010.

Optymalizacja strukturalna samolotu w układzie skrzydeł zamkniętych

Streszczenie

Układ zamkniętych skrzydeł to niekonwencjonalne rozwiązanie połączenia powierzchni

nośnych, które coraz częściej konstruktorzy starają się stosować w prototypach nowych konstrukcji.

- 10.1515/meceng-2015-0003
Downloaded from PubFactory at 08/22/2016 11:28:00PM
via free access
60 MIŁOSZ J. KALINOWSKI

Ten artykuł prezentuje przykładowy sposób realizacji optymalizacji strukturalnej struktury nośnej
skrzydeł w rozpatrywanym układzie, który może być użyteczny w trakcie projektowania wstępnego
samolotu. Na wstępie zaprezentowano metody oraz teorię wykorzystane do stworzenia algorytmu
optymalizacji. Struktura analizowana jest przy użyciu belkowego modelu MES. Optymalizacja
została przeprowadzona z wykorzystaniem połączenia metod iteracji prostych i gradientowych.
Wyniki działania algorytmu przedstawione są na prostym przypadku obliczeniowym.

- 10.1515/meceng-2015-0003
Downloaded from PubFactory at 08/22/2016 11:28:00PM
via free access