You are on page 1of 9

Aerospace Science and Technology 12 (2008) 376384 www.elsevier.

com/locate/aescte

On the stability of exible aircraft


Ilhan Tuzcu
Department of Aerospace Engineering and Mechanics, The University of Alabama, Tuscaloosa, AL 35487-0280, USA Received 12 January 2007; received in revised form 4 July 2007; accepted 17 September 2007 Available online 21 September 2007

Abstract The problem of aircraft stability has been a subject of concern since the beginnings of ight. Traditionally, aircraft stability has been treated within the connes of two separate disciplines, namely, ight dynamics and aeroelasticity. Based on some recent developments in the dynamics and control of exible aircraft, this investigation uses the system concept to provide a broader approach to aircraft stability in an attempt to bridge the gap between stability as understood in ight dynamics and stability as envisioned in aeroelasticity. To this end, stability is studied in the following four cases: 1) dynamics of whole exible aircraft using the unied formulation, 2) ight dynamics of quasi-rigid aircraft (aircraft treated as rigid), 3) aeroelasticity of exible components, such as cantilever wing, cantilever horizontal stabilizer, etc., and 4) aeroelasticity of restrained exible aircraft (aircraft xed to a point, hence, having no rigid body degrees of freedom). The paper also presents a method to address the stability of exible aircraft when the compressibility correction factor is known only at some discrete Mach numbers. 2007 Elsevier Masson SAS. All rights reserved.
Keywords: Flight dynamics; Aeroelasticity; Unied formulation; Perturbations

1. Introduction Aircraft stability has been the subject of many investigations. In most studies, the subject has been treated as two separate disciplines, namely, ight dynamics and aeroelasticity [1,3,6]. Flight dynamics is mainly concerned with the dynamics of quasi-rigid aircraft (aircraft treated as rigid) [3]. The motion is described by nonlinear ordinary differential equations of order 12, or lower, depending on assumptions. Common practice is to linearize the equations about a steady trim with the resulting equations admitting an exponential solution, so that stability can be assessed by solving the associated eigenvalue problem. On the other hand, aeroelasticity is mainly concerned with restrained exible aircraft [1,6,15]. Most common models include an airfoil section supported by springs, or a wing xed at its root, both acted upon by aerodynamic forces. The motion is described by either ordinary differential equations for the airfoil section or partial differential equations for the cantilever wing; the equations are linear or nonlinear depending on the elasticity and aerodynamics models used. Partial differential equations are transformed into sets of ordinary differential
E-mail address: ituzcu@eng.ua.edu. 1270-9638/$ see front matter 2007 Elsevier Masson SAS. All rights reserved. doi:10.1016/j.ast.2007.09.003

equations through spatial discretization. If the equations are linear, then the stability can be assessed by solving the associated eigenvalue problem. There have been very few investigations treating aircraft as they are, namely, ying exible machines. The earliest attempt goes as far back as 1929 when Frazer and Duncan [4] investigated the effect of mobility of the fuselage on utter calculations. They showed that the utter speed is larger when the rigid body degrees of freedom (dof) are included. Some important experimental and theoretical results were presented by Gaukroger [5], in which stability of two types of modes were considered: symmetric and antisymmetric modes. According to Gaukroger, in the symmetric mode, the utter characteristics depend heavily on fuselage pitching moment of inertia. With sufciently small pitching moment of inertia, utter results from the interaction of wing bending and pitch dof, which is known as body-freedom utter. The utter speed is lower than the cantilever wing utter speed and increases as the pitching moment of inertia increases, until it reaches a critical value above which the utter speed starts decreasing and slowly approaching its cantilever wing value. In the antisymmetric mode, the roll degree of freedom is the primary factor in wing utter. The utter speed might be larger than its cantilever wing

Tuzcu / Aerospace Science and Technology 12 (2008) 376384 I.

377

value depending on the roll moment of inertia. Also worthy of mention is the study by Milne [10] who derived the linearized equations of motion about a steady state by assuming small elastic and rigid body motions. The equations are used to study the longitudinal stability of the aircraft. An extensive report by Dusto et al. [2] resulting in a computer program known as FLEXSTAB, integrates exible body mechanics with a low-frequency aerodynamics. The equations are expressed in terms of steady perturbations about a reference motion and are used to determine dynamic stability. The major concern is that the structural dynamics formulation is in terms of mean axes while the aerodynamics is in terms of a different set of axes, namely, uid axes, and there is no indication of any coordinate transformation from one set of axes to the other. A fundamental study of the stability of forward swept wing aircraft is presented by Weisshaar and Zeiler [14]. A simple linear model of three dof including pitch, plunge and wing bending is used to calculate utter speeds and frequencies for various aircraft congurations. They concluded that body-freedom utter and aircraft aeroelastic divergence, not wing divergence, are primary aeroelastic instabilities experienced by forward swept wing aircraft. The instabilities are shown to occur close to the wing divergence speed, but depend upon the aircraft geometry and mass and stiffness distributions. A study by Meirovitch and Seitz [8] is concerned with the aeroelastic tailoring of a structural model consisting of a low aspect ratio swept wing made of composite materials and a rigid fuselage. The elastic wing model includes shear deformations and is attached to a rigid fuselage capable of plunge and pitch. The utter dynamic pressure and the frequency are given in terms of tailoring ply angle. A comprehensive study of nonlinear aeroelasticity of aircraft is presented in [11]. The model developed in this paper accounts for rigid body degrees of freedom and nonlinear elasticity, although, numerical results are given for cantilever wing only. The results include the effect of the nonlinearities on the utter speed and in limiting the amplitude of oscillations once an instability is encountered, i.e. existence of limit cycle oscillations (LCOs). [12] uses the model developed in [11] to study nonlinear aeroelasticity and ight dynamics of highaltitude long-endurance aircraft. The paper presents root-locus of the eigenvalues of the complete aircraft for a range of ight speeds. However, the paper does not study the divergence and utter of the complete aircraft. A fairly inclusive model of exible aircraft is given in [13]. The paper focuses on stability of a large civil transport aircraft about a reference ight, and how the eigenvalues are affected by the changes in reference roll angle, angle of attack and elevator deection. A newly developed formulation presented in a series of papers contains a rigorous and comprehensive study of dynamics and control of exible aircraft [7,9]. It integrates seamlessly in a single, consistent mathematical formulation all the necessary materials from the pertinent disciplines, namely, analytical dynamics, structural dynamics, aerodynamics, and controls. The unied formulation is based on fundamental principles and incorporates in a natural manner both the rigid-body motions and the elastic deformations, and the couplings thereof. The formulation can be used for various purposes, such as stability

analysis, simulation of aircraft motion during maneuvers, response to external excitations and real-time control of exible aircraft. The present paper attempts to paint a complete picture of the stability of exible aircraft. This is achieved by comparing results obtained by means of the unied formulation with those obtained by other approaches. To this end, stability is studied in the following four cases: 1) dynamics of whole exible aircraft using the unied formulation, 2) ight dynamics of quasi-rigid aircraft, 3) aeroelasticity of exible components, such as cantilever wing, cantilever horizontal stabilizer, etc., and 4) aeroelasticity of restrained exible aircraft. The results obtained in each of the four cases are based on the same aircraft conguration and structural and aerodynamics models. 2. The unied formulation Aircraft can be modeled as exible multibody systems, where the bodies can be identied as the fuselage (f ), wing (w) and empennage (e). To describe the motion of the aircraft, we attach sets of body axes xi yi zi , with origins Oi (i = f, w, e) to the undeformed bodies and regard the body axes attached to the fuselage, xf yf zf , as the body axes of the aircraft. Then, the motion can be expressed in terms of three translations and three rotations of the fuselage body axes and elastic deformations of exible components relative to their respective body axes. The equations of motion for exible aircraft of the type shown in Fig. 1 were derived by Meirovitch and Tuzcu [7,9]. The derivation of the equations require only expressions for the Lagrangian and the virtual work of the forces acting on the aircraft. The resulting equations include ordinary differential equations for the translations and rotations of xf yf zf and partial differential equations along with appropriate boundary conditions for the bending and torsional deformations of the exible components relative to the corresponding body axes. For practical reasons, the distributed variables must be discretized in space, which amounts to introducing the expansions ui (ri , t) = Ui (ri )qi (t), i (ri , t) = i (ri ) i (t), i = f, w, e (1)

Fig. 1. Flexible aircraft model.

378

Tuzcu / Aerospace Science and Technology 12 (2008) 376384 I.

where ri (xi , yi , zi ) is a radius vector from Oi to a typical point on body i , ui and i are elastic bending and torsional displacement vectors, respectively, Ui and i are matrices of shape functions and qi (t) and i (t) are corresponding vectors of generalized coordinates. The discrete equations of motion can be derived by introducing Eqs. (1) into the expressions for the Lagrangian and the virtual work directly. From [7,9], the state equations have the matrix form = C T V, R = E 1 , i = i , i = si , q V = p pV + F, = V pV p p + M, i = T / qi Ki qi Ci si + Qi , p i = Ki i Ci i + i , i = f, w, e where V = [u v w ]T and = [p q r ]T are the vectors of trans and skew symlational and angular velocities of xf yf zf , V pV = V p V metric matrices derived from V and such that V and p = p , R = [Xf Yf Zf ]T is the position vector of the origin Of of xf yf zf relative to the inertial axes XI YI ZI , = [ ]T is a symbolic vector of Eulerian angles between xf yf zf and XI YI ZI , C = C( ) is a matrix of direction cosines between xf yf zf and XI YI ZI , E = E( ) matrix relating Eulerian velocities to angular velocities, F = [X Y Z ]T and M = [L M N ]T are resultants of gravity, aerodynamic, propulsion and control force and moment vectors acting on the whole aircraft in terms of fuselage body axes components, Ki , Ki and Ci , Ci are stiffness and damping matrices for body i , respectively, Qi and i generalized force vectors and, nally, pV = T T , p = , V T T pi = , i = , i = f, w, e si i

in which the zero-order quantities, denoted by the overbar, are at least one order of magnitude larger than the rst-order quantities, denoted by the overhat. The elastic deformations are assumed to be of rst order by denition. Inserting Eqs. (5) into Eqs. (4) and separating the different orders of magnitude, the zero-order momenta-velocities relation can be written as [7,9] +S V = mV , p +J V = S p (6) and J are the rst and secwhere m is the total mass and S ond moments of inertia of the rigid aircraft, and hence, they are zero-order quantities. Similarly, from [7,9], the rst-order momenta-velocities relation is v v =M +M p (7) is the mass matrix for the quasi-rigid aircraft, M is the where M and v are the zero-order and rst-order perturbation about it, v is the rst-order momentum vector. The velocity vectors and p zero-order and the rst-order equations of motion can be obtained by inserting Eqs. (5) into Eqs. (2), separating the different orders of magnitude and using Eqs. (6) and (7) to eliminate the velocities. The zero-order equations of motion can be written in the compact state-space form (t) = f x (t), u(t) x
T

(2)

(8)

T p T T T = [R where x ] is the zero-order state vector, V p T u = [FE a e r ] is the control vector, in which FE , a , e and r are the engine thrust and the aileron, elevator and rudder angles and f is a nonlinear function of the state and control vectors; they represent quasi-rigid ight dynamics equations [9]. Moreover, the rst-order perturbation state equations of motion have the matrix form

(3)

(t) = A(t)x (t) x

(9)

are momentum vectors, in which T is the kinetic energy. Note that Eqs. (2) include both the velocities and momenta and, hence, must be considered in conjunction with the momentavelocities relation p = T / v = M v where p = [pT V T T T T sT e f w e ] pT pT f pT w pT e T f T w
T T e] ,

(4) v = [VT T sT f

sT w are momenta and velocity vectors, respectively, and M is the aircraft mass matrix whose expression is given in [7]. Expressions for F, M, Qi and i in terms of the actual distributed forces are also given in [7]. The state equations, Eqs. (2), represent a set of nonlinear differential equations whose order depends on the number of elastic dof of the model. The stability of such a system can be addressed by adopting a perturbation approach, which amounts to assuming that the quantities associated with the aircraft rigid body motions can be expressed as the sum of a large part, called the zero-order part, and a small part, called the rst order part, as follows: + +R , = R=R (5) +V , = + V=V

T T p T T T qT . . . T p T = [R T where x e V p e ] is the rst f ... f order state vector and A(t) = A[ x(t), u(t)] is a coefcient matrix. Eqs. (9) represent expanded aeroelasticity equations [9], in the sense that they include automatically all the aircraft rigidbody motions and can accommodate any time-varying solution (t) of Eq. (8). x The zero-order equations, or the ight dynamics equations, given by Eq. (8), are nonlinear and of order 12 at most. The rstorder equations, or the perturbation equations, given by Eq. (9) are linear and tend to be of high order. Moreover, we observe that the coefcient matrix A depends on the zero-order state (t) and control vector u(t), representing a given maneuver. If x x and u are constant, then the system of Eq. (9) is time-invariant, and u depend on time, then the system is time-varying. and if x The control vector u can be set so as to permit any given aircraft maneuver. For steady cases, such as steady level ight and level steady turn maneuver, this process is known as trimming the aircraft, which involves choosing rst the parameters of a ight, namely, speed (forward velocity), altitude and either radius or angular velocity of turn and nding the state and the control vectors that satisfy Eq. (8). However, when the aircraft executes a chosen maneuver, various distur (t), bances, such as gust can give rise to the perturbation x

Tuzcu / Aerospace Science and Technology 12 (2008) 376384 I.

379

which dies out if the rst-order system is asymptotically stable, or grows without bounds if unstable. Hence, the stability of the exible aircraft in a given maneuver can be addressed by examining the behavior of the rst-order system. When A is time-invariant, the rst-order system is asymptotically stable if all the eigenvalues of A have negative real parts. Two types of stability of exible aircraft are of interest, (0) namely, static and dynamic stability. When the initial state x is nonzero, the aircraft will generate initial forces and moments (gravitational and aerodynamic). These forces and moments are (t) to the equilibrium x =0 restoring type if they tend to take x (t) away from the equiand divergent type if they tend to take x librium. For a sufciently small ight speed, the forces and moments are restoring type and the aircraft is statically stable. However, there may exist a range of larger aircraft speed (t) for which the resulting initial forces and moments cause x to diverge from the equilibrium. If this is the case, then there (0) = 0 such that must exist a ight speed and an initial state x (0) = 0, which can be satised if and only if the determiAx nant of A is zero. Note that this initial state must be equal to an eigenvector of A corresponding to a zero eigenvalue. The ight speed for which |A| = 0 is known as divergence speed. It is important to note that for any ight speed the aircraft does not generate any force in response to the rigid body displacements R and the yaw displacement . These are all reected in four zero eigenvalues of A, and hence, the determinant of A is zero even when the speed of the aircraft is lower than the divergence speed. Therefore, the divergence speed is the lowest speed of the aircraft at which A has ve zero eigenvalues. Static stability does not really imply dynamic stability. In other words, even when the aircraft generates restoring forces (0), the dynamic and moments at t = 0 due to the initial state x forces and moments immediately after t = 0 could destabilize the aircraft. One type of dynamic instability is utter in which the aerodynamic forces result in negative damping destabilizing the aircraft. Similarly to the case of divergence, utter occurs when the speed of the aircraft reaches a value known as utter speed. This is the lowest speed for which A has at least one purely imaginary complex eigenvalue. In conclusion, divergence exists if, as the ight speed increases, any eigenvalue of A crosses the origin of the complex plane from left to right. The divergence speed is the speed of the aircraft at which the eigenvalue in question is exactly equal to zero. Similarly, utter exists if any eigenvalue crosses the imaginary axis from left to right. The utter speed is the speed of the aircraft at which the eigenvalue is purely imaginary. 3. Flight dynamics of the quasi-rigid aircraft We regard the exible aircraft shown in Fig. 1 as rigid and refer to it as quasi-rigid. The equations of motion can be obtained from Eqs. (2) by retaining the parts corresponding to the rigid-body degrees of freedom and setting all the parts corre-

sponding to the elastic degrees of freedom identically equal to zero. The resulting equations are = C T V, R V = p pV = V pV p = E 1 , + F, p + M (10)

By analogy with Eqs. (6), the relations between the velocities and the momenta are pV = mV + S T , p = S V + J (11)

where S and J are the rst and second moments of inertia for and J dened the quasi-rigid aircraft, and are equivalent to S earlier. Using these relations to eliminate the momenta, we can express the equations of motion of the quasi-rigid aircraft in the compact state-space form (t) = fQR x(t), u(t) x (12)

where x = [RT T VT T ]T is the state vector of order 12, u is the control vector dened earlier and fQR is a nonlinear function of the state and control vectors. The control vector u can be chosen so that the aircraft executes a desired maneuver. However, various disturbances cause the aircraft to undergo perturbations from the desired ight path. To address the stability of the quasi-rigid aircraft, we again +x employ a perturbation approach and write the state as x = x is one order of magnitude smaller than x . Substitutwhere x ing this into Eq. (12), separating terms of different orders of magnitude and ignoring the higher-order terms, we obtain the zero-order state equation (t), u(t) (t) = fQR x x and the rst-order perturbation state equation (t) = AQR (t)x (t) x where (t), u(t) = AQR (t) = AQR x fQR x (15)
x= x

(13)

(14)

The zero-order equation, Eq. (13), is in general nonlinear. Note that Eqs. (8) and (13) are equivalent. The only difference is that Eq. (8) is in terms of momenta while Eq. (13) is in terms of and u enter into the rst-order equation as velocities. Because x and u are inputs, Eq. (14) is linear and time-invariant if both x constant, and time-varying otherwise. As soon as a zero-order state representing a desired maneuver is chosen, the required control input enabling the maneuver can be obtained by solving Eq. (13) for u(t). Note that the quasi-rigid aircraft represents an underactuated system in the sense that the number of control inputs, which is typically four, is smaller than the number of dof. However, four control inputs are sufcient for designing most common maneuvers, such as steady level ight, steady turn and pitch maneuver. As in the case of whole exible aircraft, the stability of quasi-rigid aircraft for a given maneuver can be treated by considering the perturbation dynamics given by Eq. (14). When AQR is constant, the rst-order system is asymptotically stable if the eigenvalues of AQR have negative real parts. However,

380

Tuzcu / Aerospace Science and Technology 12 (2008) 376384 I.

we recall from the preceding section that AQR also has four zero eigenvalues. The rest are the eight eigenvalues representing typically the phugoid, short-period, spiral, rolling convergence and Dutch roll modes [3]. Note also that the existence of any positive real eigenvalue clearly indicates static instability [3]. 4. Aeroelasticity of exible components We consider cantilever models of the exible components including the fuselage, wing and empennage. The wing and the horizontal stabilizer are assumed to be xed at the mid-span, resulting in right-half and left-half cantilever wing and horizontal stabilizer, respectively. The fuselage is assumed to be xed at Of resulting in fore and aft cantilever parts of the fuselage. The vertical stabilizer is assumed to be xed at the root. Equations of motion for each component can be obtained from Eqs. (2) by retaining the elastic equations of motion for a given component and setting the elastic variables corresponding to the other components equal to zero, where the control vector u(t) and the rigid body variables are included as inputs into the equations. Hence, the equations of motion for component i take the form i = i , i = si , q i = T / qi Ki qi Ci si + Qi , p i = Ki i Ci i + i (16)

and of order 2n, where n is the number of elastic dof of the whole aircraft. Considering again the relations pi = Mi si and i = Mi i , the equations of motion can be cast in the compact matrix form (t) = ARA x(t) x (19)

T T T T where x = [qT f . . . e sf . . . e ] is the corresponding state vector and ARA is a constant coefcient matrix. The motion of the restrained aircraft consists of stable elastic vibration for the most part. However, under certain circumstances the aerodynamic forces can cause instability in the form of divergence or utter. The restrained aircraft is asymptotically stable if all the eigenvalues of ARA have negative real parts.

6. Application to a business jet As a numerical example, we consider the aircraft shown in Fig. 1 and assume a structural model as depicted in Fig. 2 of [9], in which all the exible members consists of beams in bending and torsion. In particular, the fuselage is modeled as a fore part and an aft part, both cantilevered at Of and undergoing bending vibrations in the yf and zf directions and torsion about xf . The wing consist of a right-half wing and a left-half wing modeled as cantilever beams undergoing bending in the zw direction and torsion about xw . Finally, the empennage consists of a vertical stabilizer and a right-half and left-half horizontal stabilizer, all modeled as cantilever beams undergoing bending and torsion. Each cantilever beam is assumed to be discretized by the Galerkin method using two shape functions per displacement component. For bending the shape functions are the eigenfunctions of a uniform cantilever beam and for torsion the eigenfunctions of a uniform clamped-free shaft. The model considered here is of an actual business jet made available by an aircraft manufacturer. The data regarding the geometry, the mass and stiffness distributions and the aerodynamic coefcients of the aircraft are as given in [7]. The matrices C and E and the computed stiffness matrices for the individual components, Ki and Ki (i = f, w, e ), are also given in [7]. We assume that the structural damping matrices are proportional to the stiffness matrices with the each proportionality constant being equal to 2 / where is the damping factor and is the lowest natural frequency of the respective component. For this numerical example, we assume that = 0.5% for each exible component.

where all quantities are as dened in Eqs. (2). Eqs. (16) are linear and of order 2ni where ni is the number of elastic degrees of freedom of component i . Using the momenta-velocities relations pi = Mi si , i = Mi i to eliminate the momenta, where Mi and Mi are the corresponding mass matrices, the resulting equations of motion can be cast in the compact matrix form i (t) = Ai xi (t) x [qT i T i sT i
T T i ]

(17)

is the corresponding state vector where xi = and Ai is a constant coefcient matrix. The exible component i is stable if the eigenvalue of the coefcient matrix Ai have negative real parts. In the case of steady level ight with sufciently small ight speed, all of the eigenvalues of Ai (i = f, w, e) have negative real parts. Larger ight speeds may result in negative stiffness, and hence in divergence, and/or in negative damping, and hence in utter, as explained earlier. 5. Aeroelasticity of restrained exible aircraft In this case, the exible aircraft is assumed to be xed at Of , so that it is prevented from undergoing rigid-body motions, except for steady longitudinal translation. The equations of motion for this case can again be obtained from Eqs. (2) by eliminating the rigid-body dof; the result is i = i , i = si , q i = T / qi Ki qi Ci si + Qi , p i = Ki i Ci i + i , i = f, w, e (18)

where all quantities are as dened in Eqs. (2). Note that, even though Eqs. (18) do not contain any rigid-body dof, the rigidbody variables and the control vector for a given steady maneuver enter into the equations as inputs. Eqs. (18) are linear

Fig. 2. Compressibility correction factor.

Tuzcu / Aerospace Science and Technology 12 (2008) 376384 I.

381

The maneuver of interest in this numerical example is the steady level ight, which is by far the most common case. The rst task is to obtain a trim state for a chosen steady level ight, which can be done by considering Eq. (8) (or equivalently Eq. (13)). Note that the quasi-rigid aircraft has 6 rigid body dof, so that Eq. (8) is of order 12. In this case, the zeroorder variables have the form 0 V = =0 = 0 , , R 0 0 (20) V cos =C 0 =V =0 V , 0 0 sin where V is the forward velocity of the aircraft (in the XI direction). The trim state can be obtained by substituting Eqs. (20) into Eq. (8) and solving the resulting equations for the existing unknowns (whether they are states or control inputs). Of course, this requires that the aerodynamics be explicitly known, which in turn requires that we choose the ight speed V and the altitude h. The numerical model also includes the lift-curve slopes and the control effectiveness coefcient for some number of strips on the fuselage, wing and empennage lifting surfaces [7]. The compressibility effect is included through a compressibility correction factor, fc , given for some discrete values of subsonic Mach numbers ranging from M = 0.1 to M = 0.85, as shown in Fig. 2. fc is assumed to vary linearly between two consecutive discrete Mach numbers. Also, note that fc for a given altitude h can be expressed as a function of V . In the remainder of the paper, we refer to the curve fc versus V as correction curve. In general, the distributed aerodynamic forces include circulatory and noncirculatory terms and the circulatory terms are the more signicant of the two [6]. According to the ModiedStrip theory by Yates [15], the effect of the unsteadiness of the oscillatory motion on the circulatory terms can be approximated by modifying the lift-curve slope Cl by Cl C(k) for each strip where C(k) is a complex-valued function, known as the Theodorsen function, of k = b/V , the nondimensional reduced frequency, in which V is the freestream velocity, frequency of oscillation and b the semicord [1,6]. Note that aerodynamics for C(k) = 1 is known as quasi-steady aerodynamics [6]. In this numerical example, we assume that only the circulatory terms are signicant and ignore the noncirculatory ones. As indicated in [15], the Modied-Strip theory yields more accurate results for wing of higher aspect ratios and lower swept angles. The wing of the aircraft considered in this paper has a large aspect ratio (AR = 7.3) and low swept angle (= 1.41 ) and the theory is certainly suitable for this wing. In the remainder of the paper quasi-steady and unsteady aerodynamics are labeled as QS and US, respectively. = 0, L = 0 and N = 0 are satised if The trim equations Y = 0, Z = 0 and a = 0 and r = 0. The remaining equations X = 0 can be satised by choosing , FE and e . The nonzero M variables entering as inputs into the perturbation equations, , u , w , Eqs. (9), (14), (17) and (19), are = V cos = V sin FE and e , and they are all constant. On the other hand, the only f = Vt time-varying state, namely, the forward displacement X

Fig. 3. Eigenvalues of AQR , h = 0 m, M = 0.5 to 0.85.

does not appear explicitly in the equations. Hence, the perturbation equations, Eqs. (9), (14), (17) and (19), are time-invariant, and their stability can be assessed by examining the system eigenvalues, as discussed earlier. To determine the stability of the quasi-rigid aircraft, we compute the eigenvalues i (i = 1, 2, . . . , 12) of AQR for various altitudes and discrete values of M from 0.5 to 0.85. The locus of the eigenvalues for h = 0 m (sea level) is shown in Fig. 3. As can be seen in the gure, AQR is unstable because it has four zero and one positive real eigenvalues, although the positive real eigenvalue drops from = 0.00197 to = 0.00036 as M increases from 0.5 to 0.85. The locus of the eigenvalues for h = 1524 m (= 5000 ft), h = 3048 m and h = 4572 m are similar to the one shown in Fig. 3; they are not given here for brevity. With the structural model dened in the beginning of this section, the aircraft has 32 elastic dof. Hence, the restrained aircraft has 32 dof and ARA is 64 64. With the addition of the six rigid body dof, the unrestrained aircraft has 38 dof and A is 76 76. Similarly, Af for both fore and aft fuselage are 12 12 each, and Aw for the right- and left-half wing, Ae for the rightand left-half horizontal stabilizer and the vertical stabilizer are all 8 8 each. The rst 12 eigenvalues of unrestrained aircraft are very similar to those of quasi-rigid aircraft. Hence, the unrestrained aircraft has also four zero and one positive real eigenvalues for all M = 0.50.85 and the four altitudes. On the other hand, The remaining 64 eigenvalues are similar to those of the restrained aircraft and they are stable as long as the ight speed is less than both the divergence and the utter speeds. The divergence speed VD of the unrestrained aircraft for a given altitude h can be estimated by using a trial and error approach, which starts with a sufciently small initial guess for V = VD . A corresponding compressibility correction factor from the correction curve is used to generate the aerodynamic forces and moments; a trim state is obtained by solving the trim equations mentioned above and, nally, the eigenvalues of A are computed. If A has a positive real eigenvalue, other than the positive real eigenvalue mentioned in the preceding paragraph, then V > VD and V must be slightly reduced. However, if it has no positive real eigenvalue, then V < VD and V must be slightly increased. The procedure is repeated until V is as close to VD as desired. In this manner, the divergence speed VD , for which A has a zero eigenvalue in addition to the four

382

Tuzcu / Aerospace Science and Technology 12 (2008) 376384 I.

Fig. 4. Divergence at h = 0 m.

Fig. 6. Flutter at h = 0 m.

Fig. 5. Divergence at h = 1524 m.

Fig. 7. Flutter at h = 1524 m.

zero eigenvalues mentioned earlier, can be estimated with reasonable accuracy. However, the functional dependence of fc on V introduces some complexity into this process. Perhaps, it would be easier to use a slightly different approach, as follows: First choose a compressibility correction factor fc and compute the trim states and A. Next, use the trial and error approach to estimate the lowest V such that A has an additional zero eigenvalue and repeat this procedure a number of times for different choices of fc . Then, map the resulting (fc , V ) combinations and connect the neighboring points to obtain a curve representing potential points of divergence. We refer to this curve as potential divergence curve. Next, we overlay the potential divergence curve and the correction curve for the altitude chosen. The true VD is the point of intersection of these two curves. Of course, an intersection may not exist, which would clearly mean that divergence does not exist. The method outlined in the above paragraph can be used to estimate the divergence speed using the other models, including the restrained aircraft (RA) and cantilever exible components such as cantilever wing (CW), cantilever horizontal stabilizer (CHS), etc. The method can also be used for estimating utter speed VF . The only difference is that, instead of nding the potential divergence curve, we nd the potential utter curve, which is the curve fc versus lowest V such that A has at least one purely imaginary complex eigenvalue. The true utter occurs where the potential utter curve intersects the correction curve. Flutter does not exist if the two curves do not intersect. The correction curve and potential divergence curves for the unrestrained aircraft (UA), RA and CW models at h = 0 and h = 1524 m are shown in Figs. 4 and 5. The curves for h = 3048 m and h = 4572 are similar to the ones in Figs. 4

Fig. 8. Flutter at h = 3048 m.

Fig. 9. Flutter at h = 4572 m.

and 5 and are not presented in the paper. Divergence of the cantilever wing is the most critical among the individual cantilever exible components discussed in Section 4. For this reason, only the potential divergence curve for the cantilever wing model is shown in the gure. As seen in the gure, none of the potential divergence curves intersects the correction curve, so that all three models predict that divergence does not exist for h = 0 m altitude. Divergence of the aircraft may be possi-

Tuzcu / Aerospace Science and Technology 12 (2008) 376384 I.

383

Table 1 Flutter speed and frequency, and trim values UA VF [m/s] F [r/s] [ ] FE [N] e [ ] QS US QS US QS US QS US QS US QS US QS US QS US QS US QS US QS US QS US QS US QS US QS US QS US QS US QS US QS US QS US RA h=0m 235.68 216.81 231.50 250.48 0.4680 0.5944 11714 9931.5 4.2731 4.9928 h = 1524 m VF [m/s] F [r/s] [ ] FE [N] e [ ] 244.68 225.42 231.16 250.84 0.4614 0.6114 10887 9260.1 4.0629 4.9699 h = 3048 m VF [m/s] F [r/s] [ ] FE [N] e [ ] 251.15 232.12 230.91 251.11 0.4558 0.6237 9843.1 8428.7 3.8248 4.8596 h = 4572 m VF [m/s] F [ ] [ ] FE [N] e [ ] 238.82 251.36 0.6359 7619.1 4.7442 251.99 233.40 233.66 251.50 0.4995 0.6944 8465.0 7284.6 3.9107 5.0794 220.21 208.11 241.54 248.46 0.8476 0.9990 6504.8 5832.1 5.9071 6.6620 246.21 226.47 233.81 251.25 0.5042 0.6787 9465.3 8030.7 4.1499 5.1721 207.21 196.77 241.90 248.39 0.8726 0.9815 6751.4 6107.4 6.1657 6.6588 237.90 219.29 234.04 250.97 0.5097 0.6622 10299 8771.3 4.3641 5.2465 190.53 185.03 242.39 248.32 0.9067 0.9633 6662.3 6294.1 6.3688 6.6120 228.98 210.0 234.28 250.59 0.5154 0.6396 11064 9325.5 4.5690 5.2122 174.38 174.30 242.88 248.25 0.9440 0.9473 6481.7 6475.6 6.5545 6.5747 CW

m altitudes are shown in Figs. 69. The exact values of the utter speed and frequency, and the corresponding trim variables for each model and aerodynamics used are listed in Table 1. The gain in the accuracy of the results due to inclusion of more rigid body and elastic dof, and unsteady aerodynamics is clearly seen in the gures. All six potential utter curves (three for QS and three for US) intersect the correction curve for the rst three altitudes, hence predicting utter, with the utter speed for UA being the largest of the three for the each of the cases of QS and US aerodynamics. For h = 4572 m, the potential utter curve of the UA model with QS aerodynamics does not intersect the correction curve, so that this model fails to indicate the existence of utter at this altitude. The utter frequencies, on the other hand, are close to one another for the all three models, but they are higher for the US aerodynamics than QS aerodynamics. From Fig. 6 and Table 1, it is easy to see that the utter speeds obtained from the CW model with the US and the QS aerodynamics are close to one another, but this does not mean the nature of the utter in both cases are the same. In fact, utter frequencies (Table 1) and the modes (Table 2) are signicantly different in these two cases. Hence, it can be concluded for this model that the unsteadiness in the aerodynamics has more effect on the utter frequency and the mode than the utter speed. Among the all models, the UA model with US aerodynamics is the most accurate one because it includes more rigid body and elastic degrees of freedom and more accurate aerodynamics than the rest. On the other hand, the RA and CW models with US aerodynamics yields smaller utter speeds, and hence, they are more conservative. The CW model with US aerodynamics must be used if the interest is to obtain the most conservative utter speed. The models with QS aerodynamics appear to yield larger utter speeds than the corresponding models with US aerodynamics. Perhaps, one might conclude that the actual utter speed appears to be in between the speeds from the CW model with US and the UA model with QS aerodynamics. If the interest is to design a feedback control, the most accurate model, i.e. UA with US aerodynamics, must be used. This way, the controller is less likely to spillover into the unmodeled dynamics. 7. Conclusions Recent developments in the dynamics and control of exible aircraft permit a broader approach to the concept of aircraft stability. The state equations for the ight of exible aircraft are nonlinear and generally of high order. A perturbation approach permits the separation of the problem into two problems, a nonlinear one for the ight dynamics of a quasi-rigid aircraft and a linear one for the small perturbations from the ight path, where the latter includes the elastic vibration and perturbations in the rigid-body variables. The rst can be used to dene a trim state and the second to study aircraft stability, whether for ying rigid aircraft alone, for nonying elastic aircraft, or in general for ying exible aircraft. They can also be used to produce time simulations of the rigid-body and elastic variables during ight.

ble for a more exible aircraft. In fact, all three curves move to the left, and closer to the correction curve, as the exibility increases. Note that the potential divergence velocity for UA is signicantly lower than that for both the RA and CW models. We expect this trend to persist also in the case of divergence. Hence, if divergence exists, VD for the UA model is likely to be signicantly smaller than VD for both the RA and CW models. From Figs. 4 and 5 and the potential divergence curves for h = 3048 m and h = 4572 m altitudes (not shown), it can be concluded that the onset of divergence becomes less likely as the altitude increases. The potential utter curves using both QS and US aerodynamics and the correction curve for h = 0, 1524, 3048 and 4572

384

Tuzcu / Aerospace Science and Technology 12 (2008) 376384 I.

Table 2 CW utter modes CW cordnts. ei QS US 1 .0 1 .0 1 90 90 0.0176 0.0163 qw [m] 2 48.7 73.8 0.2065 0.4076 1 97.1 102.8 0.0329 0.0629 w [rad] 2 83.4 77.7

This paper attempts to reveal the fundamental trends in the stability of the exible aircraft. To this end, it examines the following cases: 1) ying exible aircraft, 2) quasi-rigid aircraft (as in ight dynamics), 3) cantilevered exible components (as in aeroelasticity) and 4) restrained (nonying) exible aircraft. Except for the case in which the aircraft executes time-varying maneuvers, the stability can be reduced to a study of the eigenvalues of the perturbation equations. For utter prediction, two types of aerodynamics are used, namely, quasi-steady and unsteady aerodynamics. The paper also presents a method for the prediction of divergence and utter, which makes it easier to reach a conclusion about the existence or nonexistence of divergence and utter. The results presented can be improved by including more elastic degrees of freedom and rening the aerodynamics. Acknowledgements The author would like to thank Prof. Leonard Meirovitch of Virginia Polytechnic Institute and State University for his extensive review of the manuscript and his useful suggestions. References
[1] R.L. Bisplinghoff, H. Ashley, Principles of Aeroelasticity, John Wiley & Sons, Inc., New York, 1962. Reprinted by Dover Publications, Inc., New York, 1975.

[2] A.R. Dusto, et al., A method for predicting the stability characteristics of an elastic airplane, vol. 1 FLEXSTAB Theoretical Description, NASA CR-114712, October 1974. [3] B. Etkin, Dynamics of Flight, third ed., John Wiley & Sons, New York, 1996. [4] R.A. Frazer, W.J. Duncan, Wing Flutter as Inuenced by the Mobility of the Fuselage, British Aeronautical Research Council, 1929, R&M 1207. [5] D.R. Gaukroger, Wing utter, AGARD Manual on Aeroelasticity, Part V, Chapter 2, 1960. [6] D.H. Hodges, G.A. Pierce, Introduction to Structural Dynamics and Aeroelasticity, Cambridge University Press, New York, 2002. [7] L. Meirovitch, I. Tuzcu, Integrated approach to the dynamics and control of maneuvering exible aircraft, NASA CR-2003-211748, June 2003. [8] L. Meirovitch, T.J. Seitz, Structural modeling for optimization of lowaspect-ratio wings, Journal of Aircraft 32 (5) (1995) 11141123. [9] L. Meirovitch, I. Tuzcu, Unied theory for the dynamics and control of maneuvering exible aircraft, AIAA Journal 42 (4) (2004) 714727. [10] R.D. Milne, Dynamics of the deformable aeroplane, Parts I and II, Her Majestys Stationery Ofce, Reports and Memoranda 3345, London, September 1962. [11] M.J. Patil, D.H. Hodges, C.E.S. Cesnik, Nonlinear aeroelastic analysis of complete aircraft in subsonic ow, Journal of Aircraft 37 (5) (2000) 1114 1123. [12] M.J. Patil, D.H. Hodges, C.E.S. Cesnik, Nonlinear aeroelasticity and ight dynamics of high-altitude long-endurance aircraft, Journal of Aircraft 38 (1) (2001) 8894. [13] N. Siepenktter, W. Alles, Stability analysis of the nonlinear dynamics of exible aircraft, Aerospace Science and Technology 9 (2) (2005) 135141. [14] T.A. Weisshaar, T.A. Zeiler, Dynamic stability of exible forward swept wing aircraft, Journal of Aircraft 20 (12) (1983) 10141020. [15] E.C. Yates, Modied-strip-analysis method for predicting wing utter at subsonic to hypersonic speeds, Journal of Aircraft 3 (1) (1966) 2529.

You might also like