DESIGN OF A HIGH ALTITUDE LONG ENDURANCE FLYING-WING SOLAR-POWERED UNMANNED AIR VEHICLE

The low-Reynolds number environment of high-altitude §ight places severe demands on the aerodynamic design and stability and control of a high altitude, long endurance (HALE) unmanned air vehicle (UAV). The aerodynamic e©ciency of a §ying-wing con¦guration makes it an attractive design option for such an application and is investigated in the present work. The proposed con¦guration has a high-aspect ratio, swept-wing planform, the wing sweep being necessary to provide an adequate moment arm for outboard longitudinal and lateral control surfaces. A design optimization framework is developed under a MATLAB environment, combining aerodynamic, structural, and stability analysis. Low-order analysis tools are employed to facilitate e©cient computations, which is important when there are multiple optimization loops for the various engineering analyses. In particular, a vortex-lattice method is used to compute the wing planform aerodynamics, coupled to a twodimensional (2D) panel method to derive aerofoil sectional characteristics. Integral boundary-layer methods are coupled to the panel method in order to predict §ow separation boundaries during the design iterations. A quasi-analytical method is adapted for application to §yingwing con¦gurations to predict the wing weight and a linear ¦nite-beam element approach is used for structural analysis of the wing-box. Stability is a particular concern in the low-density environment of high-altitude §ight for §ying-wing aircraft and so provision of adequate directional stability and control power forms part of the optimization process. At present, a modi¦ed Genetic Algorithm is used in all of the optimization loops. Each of the low-order engineering analysis tools is validated using higher-order methods to provide con¦dence in the use of these computationally-e©cient tools in the present design-optimization framework. This paper includes the results of employing the present optimization tools in the design of a HALE, §ying-wing UAV to indicate that this is a viable design con¦guration option. Progress in Flight Physics 9 (2017) 3-24 DOI: 10.1051/eucass/201709003


INTRODUCTION
High-altitude aircraft §ying in the stratosphere (around 20 30-kilometer altitude) can provide a useful platform for sensors to support a range of military and civilian surveillance tasks.These can include real-time monitoring of seismic risks or volcanic areas, early forest ¦re detection, border security surveillance, pipe-line and power-line surveys, telecommunication services, agriculture monitoring, etc. [1].By §ying at such high altitudes, the aircraft can see or cover large geographical areas at wide angles and, in addition, the altitude o¨ers some protection in terms of possible interception by hostile vehicles.Although a number of research and operational high-altitude aircraft have been developed (including the Lockheed U-2, Helios, Zephyr, and Global Hawk) [2,3], there still remain signi¦cant challenges in the design and operation of HALE aircraft, particularly, in respect to their payload capability.
To date, the majority of existing HA UAVs have been of conventional wing/ fuselage/tail/¦n con¦guration.There is currently research being undertaken into the use of Blended-Wing Body con¦gurations for UAV applications because of their perceived advantages in terms of aerodynamic and structural e©ciency [4].A true §ying-wing is perhaps the most aerodynamically-e©cient aircraft con¦guration but, to date, has not been investigated in any detail for possible application to HA UAVs.Such a con¦guration would require a moderate amount of wing sweep in order to generate stability in §ight and to provide adequate control power for manoeuvring purposes.One objective of the present research work is to investigate the design and optimization of a swept, true §ying-wing con¦guration for application to HA UAV operations.A swept, §ying-wing con-¦guration is to be studied for operation at these §ight conditions, designed to meet a particular mission requirement.Speci¦c topics to be considered will be development of a fast and accurate-as-possible multidisciplinary optimization tool able to design and optimize such a §ying wing HALE UAV.

METHODOLOGY
The methodology which is adopted in order to complete the present project will involve the following steps.
1. Develop a multidisciplinary optimization tool which should have the following characteristics: (a) computationally fast as possible with su©cient accuracy of solution; (b) employ inviscid and viscous computational §uid dynamics (quasi-threedimensional CFD); (c) include the in §uence of structural elasticity and weight prediction; and (d) evaluate stability of the vehicle.
2. Evaluate the initial design according to a mathematical model for a solar powered aircraft to meet particular mission requirements which is 17kilometer altitude, 6-month endurance (March 1 to September 1) operating at 31-newton latitude (south of Iraq) with 50-kilogram payload for early forest ¦re detection purpose.
3. Apply to the §ying-wing con¦guration for the speci¦ed mission pro¦le and ¦nd an optimal design solution.

AERODYNAMIC ANALYSIS
Low-order analysis tools are employed to facilitate e©cient computations, which is important when there are multiple optimization loops for the various engineering analyses.This module consists of two parts: a three-dimensional (3D) inviscid- §ow model (Vortex Lattice Method, VLM) and a 2D viscid/inviscid §ow solver.Additional supplementary tools are employed such as an aerofoil section geometry generator.The Tornado VLM is used to evaluate the lift force and induced drag for speci¦c wing geometry.A 2D panel method, coupled with an integral boundary-layer method is used to evaluate 2D pro¦le drag in a stripwise sense.Spanwise integration then leads to evaluation of the pro¦le drag for the entire wing (Quasi-3D Aerodynamic Solver).The computational methods are developed and written under a MATLAB environment.

Two-Dimensional Viscid/Inviscid Module
In the present application of the incompressible §ow around an aerofoil section, viscous e¨ects are important only in the boundary-layer regions adjacent to the aerofoil surface.In this region, the governing Reynolds-averaged Navier Stokes equations can be approximated by the so-called boundary-layer equations.An inviscid- §ow model can be used outside this region [5].A coupling is required between the inviscid- §ow and boundary-layer regions.The boundary-layer calculations begin at the stagnation point at the aerofoil leading edge, with separate calculations for the upper and lower surface §ows.The viscous- §ow module consists of the following components: I Laminar Flow Field: this ¦eld begins from the stagnation point near to the leading edge to the transition point from laminar to turbulent §ow ¦eld.Thwaites£ model is used to solve the momentum integral equation [5].
Additional investigation is added to predict the laminar separation which is considered as a transition point [6,7].
II Transition Point: Michel£s criterion is used to indicate the transition point from laminar to turbulent §ow ¦eld [6].The idea of this criterion is that the transition starts at speci¦c local Reynolds number depending on the pressure gradient imposed on the boundary layer by the inviscid §ow and the surface roughness [5].
III Turbulent Flow Field: after laminar separation or beyond the transition point, the boundary layer becomes turbulent.Head£s method is used to predict the turbulent boundary layer development, using a number of semiempirical correlations of experimental data in addition to the momentum integral equation [6].This model also checks whether separation is occurring; if not, the process will continue to the trailing edge.
The outputs from this calculation are momentum thickness, shape factor, skin friction coe©cient, and displacement thickness for each panel.The tool calculates the pro¦le drag (pressure and friction drag) by using the Squire Young formula for the upper and lower surface (for more detail, see [5]).

Quasi-Three-Dimensional Aerodynamic Solver
In this method, the wing induced drag is evaluated using the Tre¨tz plane analysis (VLM), whereas the pro¦le drag is evaluated using the strip theory [8].The wing geometry is divided into several 2D spanwise wing sections, and by using the e¨ective velocity and e¨ective angle of attack, the aerodynamic forces on each segment will be evaluated.Strip Method procedure is detailed in [9].The quasi-3D solver can be divided into three main steps as shown in Fig. 1.
Step One: Tornado VLM is used to evaluate the spanwise lift distribution and induced drag of the given wing geometry and §ight state condition.The spanwise lift distribution can be interpolated for each certain strip, in addition to perform other calculations which are required to the next steps for each strip, such as strip planform area, local angle of attack, and chord length c.
Step Two: In this step, the e¨ective angle of attack (α e¨) and pro¦le drag (C de¨) , for a given local lift coe©cient (from the ¦rst step) in each strip, will be calculated by using 2D aerodynamic characteristics of the wing section (aerofoil) by applying sweep theory.It means that the airspeed V ∞ and aerofoil shape and other geometric parameters that are involved in the calculation should be based on the direction perpendicular to the sweep line as shown in Fig. 2. The sweep line can be the quarter chord sweep line while the §ight speed is in subsonic range [10].So, the perpendicular airspeed V ⊥ and perpendicular chord c ⊥ will be used instead of V ∞ and c as where C aft and C front are shown in Fig. 2.
For the tapered swept wing case, the aerofoil which lies perpendicular to the sweep line, can be interpolated from its two neighboring aerofoils as shown in Fig. 2.

Figure 3
The angles and forces present in the quasi-3D method [9] The corresponding lift coe©cient C l ⊥ is found using the sweep angle and the local lift coe©cient C l , which is determined from the lift distribution calculated by the VLM.So, it can be written as Figure 3 shows the forces and angles at each wing section.The e¨ective lift force can be given as or by using lift and drag coe©cient instead of using the forces, this becomes: The e¨ective Reynolds number can be evaluated by So, the steps which are followed in order to ¦nd the e¨ective angle of attack and pro¦le drag are shown in Fig. 4.
Step Three: The wing total pro¦le drag coe©cient C Dprof is calculated in this step based on section pro¦le strips C Dprof which is calculated from the previous step as following: Figure 4 Steps of ¦nding the e¨ective angle of attack (α e¨) and pro¦le drag So, the total drag coe©cient can be determined as a sum of wing pro¦le drag coe©cient plus the wing induced drag coe©cient: This solver has been validated with two experimental cases operating at low Reynolds number which are shown in Fig. 5.The experimental data are from [11,12].The results show that this method can give a good approximation for the lift and drag coe©cients in a short time (about 15 s in CORE i5 CPU PC).

STRUCTURE ANALYSIS
High altitude aircraft generally has an extreme span length, which suggests the aircraft is very §exible; so, it is very important to employ the elastic in §uence in the design tool.A MATLAB code has been written to size the wing box and to predict its weight by using a quasi-analytical method for a given aerodynamic load distribution (with other internal load such as weight and other inboard components: propulsion system, fuel cell, solar cell, avionic system, etc.) in addition to evaluating its deformation by using linear bending and torsion theory.The wing box is modeled by means of a 3D ¦nite element beam (thin-walled beam) concentrated on the elastic axis of the wing such as shown in Fig. 6.This method is described in more detail in [11,12].The beam is discretized into elements, each element having two nodes.Each node has 6 degrees of freedom: three in translation and three in rotation.They are described by a local coordinate and then described by their rotation angles (sweep, dihedral, and twist) to transfer to the global coordinate system.Tornado Vortex Lattice is used to evaluate the lift forces in each element.
The tool starts with calculating the bending moment, shear force and torque in each element.Then, it ¦nds the required thicknesses of the panel cross section (T b and T h , see Fig. 6), under a speci¦c load factor assuming that the wing box resists all of the external force [11,12].The steps are shown in Fig. 7. Once the element cross section is sized, then the weight will be evaluated.In addition, the ribs£ weight is evaluated from an empirical formula [12].Element sizing includes spar web and equivalent skin representation, which is produced by summing the upper and lower skin panels, stringers, and the spar §ange (see Fig. 6).The equivalent upper and lower wing panels resist bending and torsional loadings.The spar webs resist the vertical shear and torsional loads.The sizing process is iterated until the inertial relief e¨ect is achieved.
The sti¨ness matrix K for each element is then determined using the geometric and mechanical properties of the elements.The node displacement U can be evaluated by the following relation [13]: where F is the external load vectors; K is the sti¨ness matrix; and U is the displacement vector.After the sizing process, the de §ection calculation will be performed which, in turn, requires the recalculation of the aerodynamic forces; so, the sizing is then reiterated until a quasi-static equilibrium between the structural and the aerodynamic forces is obtained.

STABILITY OF FLYING-WING AIRCRAFT
Several design features speci¦c to tailless §ying-wing aircraft have been introduced, or suggested in the literature, to enhance vehicle control and stability.In terms of clean con¦guration geometry, these features are summarized as below and would need to be investigated for possible application to a HALE §ying wing UAV: using re §exed aerofoil sections to achieve the pitching moment coe©cient necessary to stabilize a tailless aircraft.However, this solution may reduce the lift at a certain angles of incidence and also reduce C L max .The maximum camber location has a strong in §uence on the pitching moment and a slight e¨ect on drag polar; so, it can be used to compensate for the lift reduction resulting from the use of a re §exed aerofoil section.The spanwise twist distribution does not give enough help in terms of stability with these types of aerofoil section [14,15]; selecting a suitable combination of sweep and twist distribution to generate zero pitching moment (with a slightly-re §exed aerofoil section) [16,17]; using a suitable dihedral angle (or dihedral distribution) could add some enhancement in the lateral direction [18]; and using winglets or C-wings to reduce induced drag and add further directional stability and control [19,20].
Here, the HS 520 aerofoil section is selected for the whole wing sections because it has a very low pitching moment and can operate in the low Reynolds number environment with a good maximum lift coe©cient as shown in Fig. 8, which is given by XFLR v6 (similar to Xfoil 2D CFD).

PRECONCEPTUAL DESIGN
A solar powered HALE UAV uses only the solar irradiance which is, in turn, dependent on the hour of the day, the day of the year, the latitude, and the position of the solar cell panels.For long endurance mission, the aircraft can §y continually if the energy collected during the daytime is enough to operate the aircraft during the whole day [21].The energy and mass balance should be the starting point of the design.Motors, solar cell panels, fuel cells or batteries, and avionic system are dimensioned according to the required power but at the same time, they have weight which a¨ects the gross weight that, in turn, a¨ects the required power.There are two di¨erent approaches to achieve the conceptual design: I The discrete and iterative approach: this is based on pure estimation for the ¦rst set of components (motors, solar cell panels, fuel cell or batteries, and avionic system) and from their weight, the total weight and power required can be estimated.This power is then compared with the previous selection and this process is performed iteratively until a converging solution is found.This approach requires more computational e¨ort with good estimation models [22].
II An analytical and continuous approach: this consists of establishing all the relationships between all the components with analytical functions using their characteristics.This approach can provide directly a unique and optimal design but requires a robust mathematical model [22].Noth et al. presented the methodology of designing a small solar powered aircraft, which was used to design the ¦rst prototype of Sky-Sailor [23].In this paper, the second approach will be adopted with some modi¦cations to the Noth methodology, particularly, in the structure mass and avionic mass and power modeling.Rizzo and Frediani proposed a structure mass estimation model which was obtained by data published for the NASA prototypes [24].This function can be written in term of span length and aspect ratio to be implemented in the design tool: The mass and power of the system components can be estimated as a constant fraction of the structural mass, or the total mass, or of the power as shown in Fig. 9. Fuel cells with a lower mass to power density ratio are used instead of batteries [1,24].

Figure 9 Schematic representation of the design methodology
The total mass m of the aircraft is the sum of all the components such as: For simplicity, variables a i can be used instead of a heavy equation as shown in Fig. 9 and after rearranging, the main equation becomes: where a 1 a 11 are de¦ned in Fig. 9.This will lead to: The positive real root of the latter equation can be found for di¨erent span lengths (b) and aspect ratios (AR), but the solution should be constrained so   that the value of the solar cell area is not more than the planform area.Table 1 shows the constant parameters of the components with the other mission requirements which are assumed still constant during the design process [23].
Applying this methodology to the design led to the possible shape of the aeroplane, for given mission requirements, as illustrated in Fig. 10.
It is clear that the span corresponding to the minimum aeroplane weight is about 56 m and aspect ratio 24.The ¦nal choice guides the §ight speed, structure weight, wing area, power and mass of fuel cells, avionic system, solar cells, avionic system, and propulsion system as shown in Table 2.The initial geometry is now determined and need to be optimized to achieve the optimal §yable geometry.

OPTIMIZATION
A design optimization framework has been developed, under a MATLAB environment combining aerodynamic, structural, and stability analysis.A Canonical Genetic Algorithm has been used in the optimization process to vary the geometric variables until achieving the optimal design which is represented by the maximum or minimum ¦tness target.The optimizer code has been developed based on principle of genetic evolutionary processes as detailed in [26,27].Here, a single objective optimization process is used to ¦nd the minimum drag for a speci¦c lift coe©cient (C Lreference = 0.9) and static margin (not less than 0.05).The ¦nal con¦guration should be stable statically and trimmed at the reference lift coe©cient and at the mission altitude.Six electric motors are used and Figure 11 Optimization architecture distributed along the span in addition to storage tanks of the fuel cells.The other fuel cell components and payload are assumed to be located at root chord and can be placed to adjust the center of gravity to achieve a speci¦c lift coef-¦cient.The communication shape among the disciplinary tools is presented in Fig. 11.A suitable mesh density is used to achieve a good sensitivity with varying the geometric variables for this con¦guration (5 chordwise × 39 semispanwise panels are used for each half span or 390 panels for entire wing).Increasing the number of panels can lead to more accuracy, but at more computational cost.
The problem here is formulated for a constant area and wing span to vary the root chord, the sweep of each part of the wing (Sw 1 and Sw 2 ), their taper ratio (TR 1 and TR 2 ), length of the ¦rst partition (b 1 ), and the twist distribution of the wing (Fig. 12).A static longitudinal stability only is considered now.

OPTIMIZATION RESULTS
The results, which are obtained from the optimization, show that it can achieve a §yable §ying wing for the given mission and §ight state.The aeroplane is trimmed at 11.7 degree angle of attack, 0.905 lift coe©cient, and 0.0193 total drag coe©cient.The gradient of the pitching moment coe©cient is −0.053 and the zero lift pitching moment coe©cient is approximately 0.0536, indicating a statically stable aeroplane in the longitudinal mode.The ¦nal con¦guration is shown in Fig. 13.The sweep required to trim this aeroplane is about 8 • with the washout twist distribution shown in Fig. 14.Only the root section seems close to stall angle at cruise condition (see Fig. 8) while the washout twist distribution (see Fig. 14) can prevent stalling for the rest of the wing.Even if the tool does not predict any separation at any section at cruise condition, this would be investigated by a high-order CFD or by an experiment for further reliability in the future work.It is known that the structure de §ection has an e¨ect on the aeroplane geometry Figure 13 The ¦nal design at Cruise and Cruise o¨conditions: 1 ¡ §ight shape; and 2 ¡ undeformed shape

Figure 14
Twist distribution for §ight (1) and undeformed shapes (2) Figure 15 Lift distribution at trim condition such as twisting in addition to increasing dihedral of the wing.The optimal undeformed wing shape is selected according to its §ight shape performance while the optimization variables are for the undeformed shape.The structure behavior is depicted in Figs. 13

CONCLUDING REMARKS
This paper presents a design and optimization framework employing aerodynamics and structure in §uence for a HALE, solar-powered §ying wing UAV.By using modi¦ed low-order analysis tools that are employed to facilitate e©cient computations for the various engineering analyses, a good approximation is achieved compared to the experimental data.Applying the §ying-wing con¦guration in this tool leads to an aeroplane that can be trimmed to carry the designed payload for long endurance (6 months) at latitude 31 N and altitude 17 km.As expected, the static longitudinal stability requires the design to be swept and twisted to give a moment arm behind the neutral axis to trim the aeroplane at the reference lift coe©cient.
Future work to be undertaken includes analyzing the dynamic stability and control, using nonlinear structure behavior, using composite materials instead of a metallic wing structure, comparison between di¨erent con¦gurations, and using multiobjective optimization to cover several objective targets.

Figure 6
Figure 6 Sketch of the structural wing model and wing-box idealization

Figure 7
Figure 7 Wing-box sizing procedure

Figure 12
Figure 12 Design variable of the half wing and 14.The lift force distributions are shown in Fig. 15.

Table 1
The constant parameters of the design * Ignored in the design.

Table 2
The main aeroplane characteristics