You are on page 1of 16

EFFICIENT APPLICATION OF CFD AEROELASTIC METHODS USING COMMERCIAL SOFTWARE

Luca Cavagna, Giuseppe Quaranta, Gian Luca Ghiringhelli and Paolo Mantegazza Dipartimento di Ingegneria Aerospaziale, Politecnico di Milano, via La Masa 34, 20156 Milano, Italy email: {cavagna,quaranta,ghiringhelli,mantegazza}@aero.polimi.it Key words: Computational Fluid Dynamics, Transonic Aeroelasticity, Flutter Analysis. Abstract. Aeroelastic analyses in transonic regime require the adoption of accurate aerodynamics physical models, such as Euler or Navier-Stokes equations. To move the application of these type of analyses from a pure academical environment to an industrial one, it is necessary to show that the technology is mature enough to be implemented without using specialized pieces of software. This paper presents a numerical procedure dened to solve Fluid-Structure Interactions (FSI) for aeroelastic problems using partitioned procedures based on the adoption of black-box commercial software for the solution of each eld. A special attention is given to the eciency of the procedure, keeping in mind the high number of analyses that have to be run during the development of a new aircraft. 1 INTRODUCTION

Aeroelastic phenomena in transonic speed range may become extremely complex, because under these conditions shock waves appear and move in the ow eld as consequence of aircraft unsteady exible motions. Usually, the appearance of shock waves may cause a drop of the utter velocity, the well known transonic dip eect, which is under-predicted by classical potential methods used for unsteady aerodynamic loads description. Improvements in the utter boundary evaluation can be obtained by using more complex descriptions of the aerodynamic domain, capable of predicting shock waves in the ow, such as those based on the Euler or Navier-Stokes equations. Starting from the pioneering works of Lee-Raush and Batina 1;2 , given the fast increase in the computer performances of the last few years, the application of unsteady CFD solutions for aeroelastic problems has grown into a large and successful research eld, with applications to complete aircraft congurations 3 . However, up to now the complexity of the procedures and the high amount of specialized computational resources required for the application of these methodologies precludes them from being extensively used in industrial aeroelastic analysis 4;5 . The purpose of this paper is to show that times are mature for trying to dene procedures to solve transonic aeroelastic problems eectively in an industrial environment. To do so, we tested the possibility to create specic procedures for aeroelastic analyses using o-the-shelf software products, such as the commercially available Computational Fluid Dynamics (CFD) software FLUENT. Figure 1 shows a block diagram with all the elements needed for conducting an aeroelastic assessment with CFD. Three main elements of this block diagram have a key role in generating an ecient and robust solution procedure: the rst is the grid interpolation between the structural and the aerodynamic discretization; the second is the grid deformation which must be used in order to adapt it to the motion of the aircraft under investigation, and third is the denition of a smart 1

CFD Grid Definition

Structural Modes

Interface

Grid Movement

Coupling Algorithm Direct Integration

CFD Solver

Transfer Matrix Identification

Figure 1: Block diagram for CFD aeroelastic analyses

numerical test procedure to be used for the identication of the unsteady aerodynamic behavior in terms of a state space Reduced Order Model (ROM). For the rst point an interfacing procedure based on a mesh-free Moving Least Square method is proposed. It ensures the conservation of the momentum and energy transfer between the uid and structure and is suitable for the treatment of geometrically complex congurations, which may include not only wings, but fuselage, nacelles and so forth. The methodology gives the user an high level of freedom to achieve the required delity and smoothness of the interpolated movements, and it is highly portable since it ensures a complete independence from the details of the numerical solvers adopted. Grid motion is another task which has a great impact on the time required by CFDCSD aeroelastic simulations. To keep a good quality of the grid through the whole simulation is very daunting job. We propose here to use a simple but robust algorithm based on the modeling of the volume grid as an arbitrary elastic solid with a prescribed stiness distribution used to to preserve the grid quality even for large movements. When reasonably small linear movements are required, the grid deformation can be computed once for the whole time transient and than simply scaled using the input variation law at the dierent time steps. In this way great computational savings are achieved. Flutter assessment can be conducted analyzing a linearized model of the equations about a certain condition found by a complete nonlinear aircraft ight trim. A linear statespace model for the unsteady aeroelastic forces can be build starting form the knowledge of the frequency domain transfer matrices using the classical Pad` e approach presented in 6 7 Ref. , or the one shown in Ref. . Consequently, a certain number of numerical experiments are needed to obtain this frequency data. It will be shown how the best trade-o between the necessity to reduce the computational time and the accuracy is obtained by computing the response of the aerodynamic system to a trimmed step input signal for each elastic mode to be analyzed. Then, by means of the Fast Fourier Transform (FFT) the complete transfer matrix in the desired range of reduced frequency is obtained with a very small number of transient computations being required. A similar identication approach is adopted by Raveh 8 which uses a wite noise modal displacements input as excitation. To assess the quality of the proposed method, a comparison with the classical experimental results for the AGARD 445.6 will be shown 9 . Furthermore, as an example of 2

Figure 2: Piaggio P-180 turboprop aircraft

application to a real life industrial case, the analysis of the main wing of the Piaggio P-180 aircraft (Figure 2) is presented. P-180 is a turboprop aircraft which must be cleared for utter signicantly beyond cruise condition, where fully transonic condition are met. 2 DEFINITION OF THE LINEARIZED AEROELASTIC PROBLEM

The primary assessment necessary for the aeroelastic certication of aircraft is related to the analysis of the local linearized solutions to nd instabilities, and specically utter points. Dierently from what can be done in the ight regimes where a linear representation of the aerodynamic eld can be adopted (i.e. potential ows), in case of strong nonlinearities in the eld such has shock waves, it is necessary to assess the stability of each movement associated with each equilibrium point of the aeroelastic system 10 . Consequently, each ight conguration may potentially assume a dierent stability behavior. However, if there are not abrupt changes in the uid ow, it is reasonable to consider the linearization around a certain ight point suciently stable for being representative of the behavior also for nearby congurations (i.e. small dierences in the mass and stiness distribution, and consequently small variation in the aircraft attitude). As a consequence, to speed up the analysis, expecially when a large number of congurations needs to be tested, it is much easier to try to extract a linearized model of the unsteady aerodynamic forces from the CFD solutions instead of running for each ight condition a new nonlinear coupled numerical utter test. The result of the linearization is a ROM for aerodynamic unsteady forces. As structural model a linear modal representation of the structure is used, as it is usually done in classic aeroelastic analysis. In any case it is necessary to have a backup procedure to run the coupled nonlinear analysis in order to verify and validate key instability points obtained by using linearized models. 2.1 Creation of the time domain reduced order model

Aerodynamic forces can be modeled as a state space dynamic system which receives as input the structural displacements, velocities, and gusts, and gives the associated generalized aerodynamic forces as output. Using a suciently large modal basis to represent the structural displacements, it is possible to identify a small set of boundary movement inputs for the aerodynamic eld. In case of structural changes which cause a variation

of the modal frequencies, if the modal basis is well chosen, and possibly hybridized with appropriate static branch modes, it is possible to see the new modes as a combination of primitive modes 11 . As a consequence, the identied model can be easily adopted for parametric analysis of the aircraft stability for dierent operative conditions. It is necessary to dene a simple excitation method which requires a reasonable computational cost but permits a good identication of the principal dynamics of aerodynamic forces, remembering that the aerodynamic system is usually over-damped. Among possible input signal the classical are: sinusoidal, impulse and step. The characterization through sinusoidal input seems the most natural but it is extremely expensive in terms of computational costs, because each modal form needs to be tested for a set of imposed frequency. The other two cases, at least ideally, require just one test for each input to characterize completely the system in the whole range of frequencies of interest. Dierent tests made have shown a great sensibility to the sampling time of the numerical discrete input realizations, expecially for the impulse case, usually requiring to increment a lot the time sampling points near discontinuities. However, it is often necessary to characterize the dynamics of generalized forces only in an assigned range of reduced frequency [0, kmax ] and not in the whole range, so it is not really necessary to adopt discontinuous input signals. As a matter of fact, it seems reasonable to apply as input a trimmed step as the following one, smoothing the discontinuities q (1 cos 0 ) 0 < max , 2 (1) q ( ) = q
max

where = tV /La is the non-dimensional time, 0 = /max and max = 2/kmax . In this way only the frequencies in the range of interest are eectively excited. Results have a very good accuracy and do not incur in numerical integration problems caused by high frequency oscillations induced in the aerodynamic eld. Furthermore, the adoption of step signals allows to compute the asymptotic value for aerodynamic forces due to a change in the boundary condition, which represent an essential data for the correct evaluation of the static gains of the transfer matrix. Of course a linearity test is always necessary to decide the correct amplitude scaling of the input signal which ensures a linear behavior of aerodynamic forces. During the simulation the aerodynamic generalized forces vector w associated with modal forms is computed. Using the FFT transformation is easy to get each column of the aerodynamic transfer matrix after each step simulation as Ham (jk, M )i = F (w(, M )i ) . F (q (, M )i ) (2)

Better results in the numerical transformation, which reduce the classical Gibbs oscillations eect near discontinuities, can be obtained expressing any generic signal s as a sum of the asymptotic value s and the deciency function Ds , which is so dened as Ds (t) = s(t) s . The deciency function for the input signal can be easily computed as q (1 + cos 0 ) 0 < max , 2 Dq ( ) = 0
max

(3)

(4)

and the Fourier transform is equal to F (q (, M )i ) = As a consequence Eq. (2) becomes Ham (jk, M )i = w + jk F (Dw (, M )i ) . qi + jk F (Dq (, M )i ) (6) q + F (Dq ). jk (5)

The transfer matrix Ham obtained in this way can be used for frequency domain analysis or can be transformed in a state space time domain system using any of the techniques currently adopted (see Ref. 6;7;12 ). The result is a state space system a = Axa + Bq, x + D2 q , fa = Cxa + D0 q + D1 q (7)

which can be connected in feedback with the structural model and so used for all types of dynamic analyses and stability investigations. A special care must be taken to correctly represent the quasi-steady coecients which are contained in the matrices D0 , D1 and D2 . They represent the system behavior at very low frequency, which corresponds to the behavior of the excited time responses near the tail of the simulations, while the system is reaching the stationary values. It must be stressed that the knowledge of the asymptotic values of the step response allows a correct static residualization of generalized forces, which is an essential ingredient for a correct comprehensive modeling of the whole dynamics of a deformable aircraft. 2.2 Direct integration of the coupled problem

The direct time integration can be easily implemented using a partitioned loosely coupled algorithm. Both systems, the structural and the aerodynamic are integrated using an implicit algorithm, leaving the time step size decision just to accuracy and sampling issues and not to numerical stability issues. For the aerodynamics, the implicit backward Euler integrator is imposed by FLUENT as the only implicit solver usable for ALE solutions. The structure is instead represented by a modal description, so the contained frequency spectrum is perfectly known and no numerical higher frequencies are present in the model. The classical partitioned scheme 13 requires for each time step: 1) a prediction of the structural displacement, which gives the new position for the structural interface; 2) the solution of the aerodynamic eld, which gives the new loads; 3) a correction of the structural time integration using the new computed loads. In this work we adopted one of the methods proposed by Giles 14 which is based on the adoption of a predictor and a corrector derived form Crank-Nicholson algorithm: Predictor q = Corrector q
(n+1)

1 I + hA 2 =

1 I hA q(n) + hp(n) , 2
1

(8) , (9)

1 I + hA 2

1 1 I hA q(n) + h p(n+1) + p(n) 2 2

where A is the state space matrix used to represent the modal structural model, q is the vector of the structural modal states (i.e. modal position and velocity), and p is the nodal 5

aerodynamic loads vector. Even though partitioned loosely coupled methods may create a net energy loss/increment during the simulation, because
n+1) ( f (p(n) x p(n+1) x s (n+1)

) dA =
B

(n+1) s (p(n) p(n+1) ) x dA = 0

(10)

Giles 14 shown that choosing a suciently small time step ensures the overall stability of the system. 3 INTERFACING DATA BETWEEN FLUID AND STRUCTURE

The adoption of a partitioned approach for the solution of Fluid-Structure Interaction problems requires the denition of an interface scheme to exchange displacements and velocities from the structural grid to the aerodynamic wet surfaces of the CFD grid and to transfer back aerodynamic forces on the structural nodes. The two models are discretized in a very dierent and often not compatible way; this is especially true in an industrial environment, where they usually come from dierent departments. Structures are represented by complex denition volumes, often very discontinuous. Their numerical representations are based on the adoption of schematic models, which have a long tradition in the aerospace industry, made by elements with very dierent topologies, such has beams, plates and solid elements, which usually are not coincident with the real geometrical representation of the aircraft (see Figure 3). It is the authors opinion that these simplied models will be used for some time to come in aerospace industry for dynamic analysis, so it is essential to be able to cope with them. On the other side, aerodynamic grid requires an exact representation of the wet surfaces, so it is necessary to make these two representations of the same aircraft compatible in order to transfer information between them. A correct and ecient interface scheme for partitioned analysis must possess all those properties: possibility to interface both non-matching surfaces or non-matching topologies; capability to deal with situations where a control point fall outside the range of the source mesh (extrapolation); exact treatment of rigid translations and rotations; capability to deal correctly with situations having a wide variation of the node density of the source mesh; independence from the numerical formulation of the Computational Fluid Dynamics (CFD) and Computational Structural Dynamics (CSD) solvers; conservation of the exchanged quantities (in particular momentum and energy); possibility to control the smoothness of the resulting surface. The last two points are essential when stability analysis have to be carried out. The comparison of spurious energy created or canceled by the interface scheme may alter the stability boundary of the system. Furthermore, when highly accurate aerodynamics models are used, such has Euler or Navier-Stokes, a non correct smoothness of the wet 6

Figure 3: Results of the interface procedure for the rst two Piaggio P180 wing out-of-plane modes

surface may cause numerical convergence problems or unphysical local instabilities of the ux. In order to guarantee the conservation between the two models, the correct strategy would be to enforce the coupling conditions only in a weak sense, through the use of simple variation principles such as that of Virtual Work. Let yf and ys be two admissible virtual displacements for each eld. Admissible means that the trace of these two elds on , which can be either a newly dened virtual interface surface or simply the surface of the uid eld f which is always present, must be equal Tr ( yf )| = Tr ( ys )| . (11)

After computing the nodal loads (Ff )i for the aerodynamic boundary grid points using the correct approximation space, the loads on the structural nodes (Fs )j , can be obtained by simply multiplying the formers by the transpose of the interpolation matrix H that connects the two grid displacements
js

( yf )i =
j =1 if

hij ( ys )j ,

(12)

(Fs )j =
i=1

hij (Ff )i .

(13)

This is a well known result, reported in almost all works about the implementation of an interface algorithm 15;16 , that ensure the balance of the energy exchanged between the uid and the structure. However, this point does not completely solve the conservation issue, because there is no conservation of the velocity transmitted from the structure to uid boundaries, so no guarantee about the conservation of the momentum transferred to uid. The problem of conservation is now shifted on the denition of the correct interpolation matrix H .

To build a conservative interpolation matrix which enforces the compatibility, Eq. (11), a weak/variational formulation can be used. The idea is to express the problem as a weighted least-square problem Minimize

(Tr ( yf )| Tr ( ys )| )2 dA.

(14)

In addition to this, additional properties can be sought, like smoothness of the resulting interpolated eld, computational eciency and some control on the interpolation error. A solution which possesses all these qualities can be obtained using the MLS technique. The origin of this approximation is connected to the surface or data reconstruction eld (see Lancaster and Salkauskas 17 , and Schaback 18 ). The problem can be mathematically stated as follows. Given a compact space Rn , the object of the analysis is the reconstruction of a function f C d () from its values f ( x1 ), f ( x2 ), . . . , f ( xN ) on scattered distinct 1, x 2, . . . , x N }. Of course, it is not necessary to derive an analytical centers X = {x expression for f ; it is sucient to have an ecient method to compute the value of f on a dierent set of centers Y = {y1 , y2 , . . . , yN }. The method should ideally have these properties: a) computational eciency; b) correct smoothness of the resulting surface; c) quality of reproduction. The rst step is to build a local approximation of f as a sum of monomial basis functions pi (x) Pd
m

= f

i=1

pi (x)ai (x) pT (x) a(x),

(15)

where m is the number of basis functions, and ai (x) are their coecients. Pd C d () is a nite dimensional space of basis functions; usually it is spanned by polynomials, but other forms can be adopted. In this case the adopted basis functions are either linear or quadratic polynomials pT (x) = (1, x, y, z )T C 1 (R3 ), (16) C 2 (R3 ). (17)

pT (x) = (1, x, y, z, x2 , xy, y 2, yz, z 2 , zx)T

The coecients ai (x) are obtained by performing a weighted least square t for the approximation 2 f ( ) f x) d( Minimize J (x) = (x x x), (18)

under the linear constraint (x) = f

pi ( x)ai (x).
i=1

(19)

This equation is completely equivalent to Eq. (14) which expresses the interface problem. The great advantage of the problem expressed in this form is that it can be localized by choosing compact support weight functions such as smooth nonnegative Radial Basis Functions (RBF). Usually the weight RBF are written as (r/ ), where delta is a scaling factor that allows to change the function support at dierent space centers. The parameter allows the user to adapt the support radius to the problem, being sure that, on the one hand, enough points are covered, and, on the other hand, far away points have no inuence. More details on the implementation of this interface scheme together with few application results may be found in Ref. 19 . 8

EFFICIENT GRID DEFORMATION

To correctly represent the structural deformation of the aircraft, the CFD computational grid must be modied at each time step in order to be compatible with the structural deformation. Using an Arbitrary Lagrangian Eulerian formulation of the uid equation 20 , it is possible to keep into account any possible movement of grid nodes. Consequently, by simply deforming the uid grid, it is possible to solve the problem easily without generating a new grid at each time step. Of course, the deformed grid must follow the structural movements, but, at the same time, must keep a good quality in order to avoid any numerical problem during the simulation. There is a large literature about methods to obtain compatible deformed grid. Batina 21 introduced the elastic analogy representing each side of the grid as a spring with a nonlinear stiness proportional to the side length. To avoid the occurrence of invalid elements with negative volumes during the simulation, Degand and Farhat 22 introduced torsional springs at each vertex, making the algorithm even most expensive in terms of computational costs. If not treated in the right way, the deformation problem may become one of the most expensive task of this types of simulations. In this work a dierent way is followed. The rst goal sought is the overall computational eciency, so we tried to avoid any nonlinear model for grid deformation. Also in this case an elastic analogy is exploited, but the grid is represented as a linear elastic continuum with a local Young modulus proportional to the minimal dimension of each element following a law of this type Eel = 1 min xj xk

(20)

j,k el

A Poisson coecient [0; 0.35] is chosen in order to avoid numerical bad conditioning of the problem. This distribution of stiness allows to relieve the eects of structural deformations from inner small elements near the aircraft surface leaving the burden on outer larger elements, which can be deformed without large distortions. This method works well for any element shape and also for Navier-Stokes hybrid meshes made by tetrahedrons and hexahedrons. An example of local grid deformation for very large structural deformation is shown in Figure 4. In this case the AGARD 445.6 wing is moved following the pattern of the fth mode. In the shown cases the grid is still valid, in the sense that no element volume is negative and both the overall and local quality is kept almost constant. In fact, the structural analogy gives to the user a large freedom in choosing the material constitutive properties to rule the grid deformation behavior during the simulation. By simply choosing the structural properties it is possible to work on the quality of the grid during the deformation phase. Furthermore, the linearity of the problem allows more room for computational savings. Using a direct method for the solution of related linear algebra problems, the matrix factorization can be done once for all. The new deformed grid at each time step may be found by a simple and fast application of the backward and forward substitution steps. In the tests presented here, the structural model is always represented as a set of modal forms. Consequently, also the CFD grid deformation can be represented, exploiting the linearity of the problem, as a superposition of grid deformations computed for each modal form. In this way the grid deformation is almost a trivial task which requires very small computational resources and time.

Figure 4: CFD unstructured grid section deformation along the fth mode of AFGARD 445.6 wing

VALIDATION ON AGARD 445.6 WING

A well-known three-dimensional standard aeroelastic conguration is considered to validate the whole procedure: the AGARD 445.6 weakened wing which was tested in windtunnel at NASA Langley. The wing semispan model is made of laminated mahogany, with NACA 65A004 airfoil, a quarter-chord sweep angle of 45 deg, and an aspect ratio of 1.65 and a taper ratio of 0.66. To reduce the stiness, the wing was weakened by holes drilled through it and lled with foam. In this section we refer to the weakened model number 3 since all ow-conditions (subsonic, transonic and supersonic) were tested. Further data concerning the models, test setups and conditions are reported in Ref 9 . The structural model is represented, by the rst four normal undamped modes. The rst two are primarily involved in the utter mechanism, identied respectively as rst bending and rst-torsional modes by a nite-element analysis (Figure 5). Since no structural damping is available for the wing model, a estimate value of modal damping coecient equal to 0.01 is applied. The aerodynamic system is based on the Euler equations representing a good compromise between accuracy in description of the transonic eects and computational burden. Moreover, no information is available on the experimental transition location. A C-H type mesh of 124.160 cells is built, extending 7 root chord lengths from the wing to the upstream and downstream boundaries and 3 semi-span lengths from the tip to the side boundary. Stability behavior is determined at null incidence for three dierent Mach number conditions, 0.678, 0.96 and 1.141, in order to assess the quality of the dened procedure in all ow-conditions, subsonic, transonic and weakly supersonic with a detached curved shock bump. The airfoil section is symmetric, so no static aeroelastic trimming procedure for wing deections is required. The computational of each column of the transfer matrix requires about 2 hours with a single Pentium IV 2.8 MHz. Flutter results for the subsonic and transonic cases agree with experimental data within 1% (Figure 6). In the supersonic case, similarly to Refs. 1;23 , 20% higher utter velocity than those determined in experiment are found. Although Ref. 2 showed some improvements by a Navier-Stokes model, the degree of improvements was small compared to the still large dierence with experimental values. For this case our tests have shown that a better grid space resolution near the leading edge with inviscid model leads to 10

Figure 5: AGARD 445.6 structuredd grid deformed along rst two structural normal modes

0.6 0.55 0.5

0.7 0.65 0.6 0.55

0.45 0.4 0.35 0.3 / Experiment CFDEuler 0.5 0.6 0.7 0.8 0.9 Mach number 1 1.1 1.2

0.5 0.45 0.4 0.35 0.3


f

VF

0.25 0.2 0.4

0.25 0.2 0.4 0.5 0.6 0.7 0.8 0.9 Mach number 1

Experiment CFDEuler 1.1 1.2

Figure 6: Flutter speed index and frequency ratio for the AGARD 445.6 wing

improvements in correlation with experiments. The ow behind the detached shock bump needs to be correctly evaluated because has an high inuence on pressure loads on the whole wing behind. For this reason we suppose Navier-Stokes improvements are due to the mandatory better grid resolution in order to correctly model the viscous layer. Future works will presents results of this research. 5.1 Analysis by means of direct integration

Unlike linearized utter analysis where motion is prescribed using a trimmed step input law, direct integration is used to study the time evolution of the aeroelastic system by a direct CSD-CFD coupled simulation. The wing motion is excited with a user-denable initial modal velocity perturbing a steady trimmed aerodynamic solution. Input amplitude is determined by maximum local incidence variation based on maximum modal displacement. As Figures 7 - 8 show, the wing starts oscillating resulting in damped, neutral or diverging vibrations. The diverging oscillations evolve very fast toward a limit cycle. This test case shows how the grid deformation routines, and the entire procedure, are robust enough to solve the problem even with large deformations, such as those met in this simulation. 6 APPLICATION TO THE PIAGGIO P-180 AVANTI WING

Piaggio P180 (Figure 2) is a twin turbo-prop pushed aircraft whose ight envelope, according to current regulations, must be clear from utter instability up to Mach 0.92, 11

0.015

0.15 0.1 0.05

0.01

Modal amplitude

Modal amplitude

0.005

0 0.05 0.1 0.15 Mode 1 Mode 2 Mode 3 Mode 4 0 4 8 12 16 20 24 28 Structural adimensional time 32 36 40

0.005

0.01 Mode 1 Mode 2 0.015 0 4 8 12 16 Structural adimensional time 20 24

0.2

Figure 7: Modal amplitudes history for the AGARD 445.6 wing, M = 0.678, VF = 0.34

Figure 8: Modal amplitudes history for the AGARD 445.6 wing using 4 modes, M = 0.678, VF = 0.50

far beyond Mach cruise approximately set to Mach 0.7. Cruise conditions are signicantly marked by transonic eects manifesting as a shock on the aft portion of the lifting surfaces. P180s layout is not conventional since three lifting surfaces are used in order to reduce induced drag. Moreover its wing has a full-immersed nacelle giving rise to phenomena which are rather complex to be analytically modeled: engine-inlet and outlet, pushing props inuence on local ow, body-interferences with the surrounding wing. In this work,we present the rst steps taken to model the P180 as an aeroelastic system, starting from the isolated wing model. In this way it is possible to lay out the foundations for a full-plane procedure. 6.1 Model description

A nite-element model of the wing-box is used to describe P180s structural dynamic system. First two undamped normal modes, respectively the rst bending and rst torsional, are used for wing utter analysis after considering dierent simplied DoubletLattice Models (DLM). The aerodynamic model is based on Euler equations. Given the geometric complexity, an unstructured grid consisting of 274.740 cells is built to model the ow-eld around the wing. Rear-pushing propeller aerodynamic over-velocity eects are neglected since they are small and considered not fundamental for the analysis of utter mechanism. Correct aerodynamic propeller-modeling, besides adding work terms in the utter equation considering thrust as a follower force, will be the object of future works. In order to add respectively inlet and outlet eects, two iterative procedures are used while solving reference steady aerodynamic solution: the former changes local static pressure in order to get a xed mass ow for the engine; the latter changes local total temperature to get a xed thrust. Good inlet modeling turned out to improve steady aerodynamics analysis convergence. 6.2 Aerodynamic validation

Various steady cases are investigated in order to validate the correct pressure loads of the CFD model, neglecting aeroelastic static deections. Figure 9 shows good agreement with transonic wind-tunnel tests. Despite using an inviscid model, shock locations are well dened and so is the overall pressure load, especially if we consider that the reference measured prole section is particularly close to the nacelle. An exception is represented 12

by the discrepancy between the last but three sample point and CFD pressure coecient: ap gap has been closed in grid building phase in order to avoid unuseful increase in cell number and because local solution nearby control surfaces are not considered important for a overall stability. Lift curve slope agrees with experimental wind-tunnel tests. A
1.5 1.25 1 0.75 0.5 Cp 0.25 0 0.25 0.5 0.75 1

1.5 1.25 1 0.75 0.5 Cp 0.25 0 0.25 0.5 0.75


1

CFDEuler Experiment 0 0.1 0.2 0.3 0.4 0.5 x/c 0.6 0.7 0.8 0.9

CFDEuler Experiment 0 0.1 0.2 0.3 0.4 0.5 x/c 0.6 0.7 0.8 0.9 1

Figure 9: Comparison of pressure distributions, y/b=0.2, M = 0.7, = 2.259 and 4.383 respectively

rst dynamic example is shown in Figure 11 representing aerodynamic generalized force coecient due to plunge motion. The experimental steady rigid value is conrmed by the result extrapolated from numerical tests.
0.75 0.5 0.25 0 0.25 0.5 0.75 1 Real CFD Imaginary CFD Real DLM Imaginary DLM 0 0.1 0.2 0.3 Reduced frequency k 0.4 0.5

Figure 10: Pressure distributions on the wing, M = 0.7, = 0

6.3

Aeroelastic utter analysis

Figure 12 shows the computed coecient of the aerodynamic transfer matrix at Mach 0.7, comparing them with the linear results of DLM. Discrepancy among coecients grows as Mach number increases. The CFD results are compared with two linear models: a simple model of the lifting surface with equivalent panels in the nacelle zone; and one using slender and interference bodies implemented in MSC-NASTRAN aeroelastic solver in order to account for wing-nacelle interactions and their eect on the pressure load. Despite a transonic ight condition is studied, DLM results agree with CFD utter procedure within 2%; no transonic dip eect is shown. Table 1 reports utter results (not Mach-aligned) at sea level for dierent ight conditions. Two dierent eects are 13

hh

[m2]

Figure 11: Generalized plunge aerodynamic forces for Piaggio P180 wing compared to DLM, M = 0.7

M VFpanels VFCF D VFbody

0.7 0.8 203.6 203.9 207.8 209.2 210.2 212.4

0.9 205.7 210.4 210.5

Table 1: Flutter velocity (m/s) at dierent Mach numbers, z = 0m

0.5 0.25 0

1 0.75 0.5

0.25 0.5 Q11 [m ]


12

0.25 [m2] 0 0.25 0.5

0.75 1 1.25 1.5 1.75 2 0 0.1 0.2 0.3 Reduced frequency k Real CFD Imaginary CFD Real DLM Imaginary DLM 0.4 0.5

0.75 1

Real CFD Imaginary CFD Real DLM Imaginary DLM 0 0.1 0.2 0.3 Reduced frequency k 0.4 0.5

0.5

1 0.75 0.5

0.25

0 Q21 [m ] [m2]
2

0.25 0 0.25 0.5

0.5 Real CFD Imaginary CFD Real DLM Imaginary DLM 0 0.1 0.2 0.3 Reduced frequency k 0.4 0.5

22

0.25

0.75

0.75 1

Real CFD Imaginary CFD Real DLM Imaginary DLM 0 0.1 0.2 0.3 Reduced frequency k 0.4 0.5

Figure 12: Generalized aerodynamic forces for Piaggio P180 wing compared to Doublet Lattice Method, M = 0.7

combined to give results which are so similar with such dierent aerodynamic models. On one side is the eect of the shock wave which cause an anti-stabilizing eect. On the other side, as can be seen from Figure 10, nacelle interference plays an important role in load distribution: besides a sharp drop in pressure load, aerodynamic pressure resultant is placed far backward from the elastic axis creating a stabilizing terms that cannot be caught with the simplied DLM geometry and that results in a delayed utter condition. The results obtained by means of the linearized analysis are conrmed by direct analysis of the coupled systems. Figure 13 shows the unstable movement of the two modal components at a velocity slightly above the utter condition for Mach 0.7. The initial excitation is automatically generated since the system is not starting from a trimmed condition for the elastic deformations. 7 CONCLUSIONS

In this paper a procedure to apply CFD solutions for aeroelastic stability evaluation has been presented. It has been displayed how, by a careful choice of the basic algorithmic elements necessary for this type of analyses, it is possible to obtain a methodology rou14

0.2 0.1 0 0.1 Modal amplitude 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 2 4 6 Structural adimensional time 8 Mode 1 Mode 2 10

Figure 13: Post utter direct simulation of P-180 at M = 0.7: modal amplitudes history

tinely applicable for analyses in an industrial environment. To further demonstrate the potential of the proposed technique, applications to complete aircraft congurations are currently under way and will be presented soon. 8 ACKNOWLEDGMENTS

The authors wish to acknowledge Piaggio Aero Industries for providing P-180 data, Sara Bozzini, Raaele Ponzini and Paolo Ramieri, members of Cilea Consortium, for the technical support with Avogadro computing cluster.

References
[1] E. M. Lee-Raush and J. T. Batina, Wing utter boundary prediction using unsteady Euler aerodynamic method. AIAA Paper 93-1422, April 1993. [2] E. M. Lee-Rausch and J. T. Batina, Wing utter computations using an aerodynamic model based on the Navier-Stokes equations, Journal of Aircraft, vol. 33, no. 6, pp. 1139 1147, 1996. [3] R. Melville, Nonlinear mechanisms of aerelastic instability for the F-16. AIAA Paper 2002-0871, January 2002. [4] R. M. Bennet and J. W. Edwards, An overview of recent developments in computational aeroelasticity, in Proceedings of the 29 th AIAA Fluid Dynamics Conference, (Albuquerque, NM), June 15-18 1998. [5] D. M. Schuster, D. D. Liu, and L. J. Huttsell, Computational aeroelasticity: Success, progress, challenge, Journal of Aircraft, vol. 40, no. 5, pp. 843856, 2003. [6] L. Morino, F. Mastroddi, F. De Troia, G. L. Ghiringhelli, and P. Mantegazza, Matrix fraction approach for nite-state aerodynamic modeling, AIAA Journal, vol. 33, no. 4, pp. 703711, 1995.

15

[7] A. Scotti, G. Quaranta, and S. Ricci, Active control of three surface wind tunnel aeroelastic demonstrator: Modelling and correlation, in International Forum on Aeroelasticity and Structural Dynamics IFASD-2005, (Munich, Germany), June 28 July 1 2005. [8] D. E. Raveh, Identication of computational-uid-dynamics based unsteady aerodynamic models for aeroelastic analysis, Journal of Aircraft, vol. 41, no. 3, pp. 620632, 2004. [9] E. C. Yates, AGARD standard aeroelastic congurations for dynamic response. I wing 445.6, R 765, AGARD, 1985. [10] A. Lyapunov, The General Problem of the Stability of Motion. Princeton University Press, 1947. [11] V. Giavotto, P. Mantegazza, L. D. Otto, M. Lucchesini, and R. Mantelli, Fast utter clearance by parameter variation, CP 354, AGARD, September 1983. [12] M. Karpel, Reduced order aeroelastic models via dynamic residualization, Journal of Aircraft, vol. 27, no. 5, pp. 449455, 1990. [13] C. Farhat and M. Lesoinne, Two ecient staggered algorithms for the serial and parallel solution of three-dimensional nonlinear transient aeroelastic problems, Computer Methods in Applied Mechanics and Engineering, vol. 182, pp. 499515, 2000. [14] M. B. Giles, Stability and accuracy of numerical boundary conditions in aeroelastic analysis, International Journal for Numerical Methods in Fluids, vol. 24, no. 8, pp. 739757, 1997. [15] N. Maman and C. Farhat, Matching uid and structure meshes for aeroelastic computations: a parallel approach, Computers & Structure, vol. 54, no. 4, pp. 779785, 1995. [16] M. J. Smith, D. H. Hodges, and C. E. Cesnik, Evaluation of computational algorithms for suitable uid-structure interactions, Journal of Aircraft, vol. 37, no. 2, pp. 282294, 2000. [17] P. Lancaster and K. Salkauskas, Surfaces generated by moving least squares methods, Mathematics of Computation, vol. 37, pp. 141158, 1981. [18] R. Schaback, Remarks on meshless local construction of surfaces, in Proceedings of IMA Mathematics of Surfaces IX Conference, (Cambridge), 2000. [19] G. Quaranta, P. Masarati, and P. Mantegazza, A conservative mesh-free approach for uidstructure interface problems, in International Conference on Computational Methods for Coupled Problems in Science and Engineering (M. Papadrakakis, E. O nate, and B. Schreer, eds.), (Santorini, Greece), CIMNE, 2005. [20] J. Donea, Arbitrary lagrangianeulerian nite element methods, in Computational Methods for Transient Analysis (T. Belytschko and T. J. Hughes, eds.), ch. 10, pp. 474516, Amsterdam, The Netherlands: Elsevier Science Publisher, 1983. [21] J. Batina, Unsteady euler airfoil solution using unstructured dynamic meshes, AIAA Journal, vol. 28, pp. 13811388, 1990. [22] C. Degand and C. Farhat, A three-dimensional torsional spring analogy method for unstructured dynamic meshes, Computers and Structures, vol. 80, pp. 305316, 2002. [23] F. Liu, J. Cai, Y. Zhu, H. M. Tsai, and A. S. Wong, Calculation of wing utter by a coupled uid-structure method, Journal of Aircraft, vol. 38, no. 2, pp. 334342, 2001.

16

You might also like