Finite State Induced Flow Models Part I: Two-Dimensional Thin Airfoil David A . Peters* Washington University, St. Louis, Missouri 63130 Swaminathan Karunamoorthyt St. Louis University, Cahokia, Illinois 62206 and Wen-Ming Cao$ Washington University, St. Louis, Missouri 63130 A new finite state aerodynamic theory is presented for incompressible, two-dimensional flow around thin airfoils. The theory is derived directly from potential flow theory with no assumptions on the time history of airfoil motions. The aerodynamic states are the coefficients of a set of induced-flow expansions. As a result, the finite state equations are hierarchical in nature and have closed-form coefficients. Therefore, the model can be taken to as many states as are dictated by the spatial texture and frequency range of interest with no intermediate numerical analysis. The set of first-order state equations is easily coupled with structure and control equations and can be exercised in the frequency or Laplace domain as well as in the time domain. Comparisons are given with Theodorsen theory, Wagner theory, and other methods. Excellent results are found with only a few states. Nomenclature [A] = matrix of acceleration coefficients, Eq. (35) [B ] = eigenvectors of [A] b = semichord, m b n = coefficients of expansion, A ppendix C C(k) = Theodorsen function c,, = vector of length A f, Eq. (31) > , = matrix, Eq. (30) d,, = vector of length N, Eq. (31) f(x) = generic function /(*) = g(rj) (^ K 1 ? ) = generic function / = identity matrix j index taken as s, a, n k = reduced frequency L = lift per unit length divided by 27rpbV 2 M = pitching moment about midchord over 2irpb 2 V 2 A /, /4 = normalized moment about quarterchord m = index 1, . . . , N N = number of inflow states n index 1, . . . , N P f) pressure differential on airfoil, nondimensional on pV 2 Q[f] = functional, Eq. (18) s = L aplace variable T n = eigenvalues of [A] t = reduced time, normalized on semichord and freestream u = x component of induced velocity, normalized on V V = freestream velocity, m/s R eceived Sept. 13, 1993; revision received May 23, 1994; accepted for publication J une 17, 1994. Copyright 1994 by the A merican I nstitute of A eronautics and A stronautics, I nc. A ll rights reserved. :!: Professor and Director, Center for C omputational Mechanics, Campus Box 1129. Fellow A I A A . t A ssociate Professor, Department of A erospace Engineering, Parks College. Member A I A A . ^ Graduate R esearch A ssistant, Department of Mechanical Engi- neering, Campus Box 1185. v = y component of induced velocity, normalized on V W(t) = Wagner function w = total induced downwash, -v Wj = coefficients of w expansion, Eq. (22) J T , y = Cartesian coordinates normalized on b Y(Z) = weighting function Z = wake variable, e ~ ^ F = total bound vorticity divided by bV F = normalized vorticity, 172 TT y = vorticity density divided by V y h = bound vorticity density jj = expansion coefficients of vorticity and velocity, Eqs. (11) and (12) y w = wake vorticity density A = elliptical determinant, A ppendix A V = L aplace operator 8(t) = impulse function e = residual error, Eq. (43) A = induced flow due to shed vorticity divided by V A , = expansion terms for A , Eq. (17) A y - = complex representation of A , f = streamwise direction p = density of air, kg/m 3 TJ = pressure expansion coefficients, Eqs. (13) and (14) 4> y = acceleration potentials, A ppendix A ^ = velocity potentials, A ppendix A Superscripts A = from acceleration term T = transpose V = from velocity term = - d/ dy ^ = d/ dt = evaluated on upper surface of airfoil Subscripts a = asymmetric potential L = lower surface 5 = symmetric potential U = upper surface 313 314 PETER S, KA R U N A MO O R T HY, A N D CA O : TWO -DI MEN SI O N A L T HI N A I R F O I L Introduction Background U N STEA DY aerodynamic models that are useful for aeroelastic analysis are usually of four general types. F irst, there are /c-type aerodynamic theories in which the motions, pressure, and induced flow undergo only simple har- monic motion, e ikt . These are often used in the V-g method but, strictly-speaking, they are accurate only at the stability boundary. A second category of model hasp-type (or L aplace- domain) aerodynamics in which the aerodynamic variables undergo exponentially growing (or decaying) harmonic mo- tion. This type of model is utilized in eigenanalysis in which an iteration on p (i.e., on s) is performed for every eigenvalue of interest. A third type of aerodynamic model is indicial, in which a Green's function is utilized with a convolution integral to give arbitrary motion response. This is useful in time- marching. In principle, these three types of aerodynamics are equivalent and can be derived from one another through L a- place and F ourier transforms. A fourth type of unsteady aerodynamics is the class of finite state models. In a precise sense, two-dimensional unsteady aerodynamics has no finite state representation, but is an infinite state process [e.g., s /,/(s) terms appear in the transfer function]. N evertheless, useful finite state approximations have been derived. There are several advantages of finite state models. F irst, finite state modeling allows one to cast the aerodynamics in the same state-space context as the structural dynamics and controls. This allows the full complement of control theory and systems theory to be brought to bear on the problem of aero-servo-elastic control and design. Second, the existence of explicit states eliminates the necessity to it- erate on solutions (as in V-g and p-k methods). I nstead, the entire solution can be obtained in one pass. Third, a state- space model is flexible in that it can be exercised in the fre- quency domain, L aplace domain, or the time domain as de- sired. There are several types of finite state models. Vortex-lattice and computational fluid dynamics (CFD) methods can be cast as finite state models with the number of states being the order of the number of lattice nodes or CF D grid points. This, however, is usually such a large number of states that con- ventional control-theory applications are precluded. I nstead, most applications of finite state models to aeroelasticity have utilized a relatively few approximate states. The disadvan- tages of this are that such states have no direct physical inter- pretation and that they cannot be systematically improved in a hierarchical manner. I t would seem, therefore, useful to have a more general finite state model. Previous Work I n 1925, Wagner published the indicial function for the lift response of a two-dimensional, flat-plate airfoil in incom- pressible flow. 1 In 1935, Theodorsen presented the lift fre- quency response for the same conditions. 2 Garrick then showed that the two were related (and mutually consistent) by means of F ourier transform. 3 The use of L aplace transform (/? -ver- sion aerodynamics) was suggested by J ones, 4 and applied to some problems by Sears. 5 J ones 4 obtained an approximate L aplace transform of the Wagner function, 6 but J ones 7 was the first to generalize the Theodorsen function formally for p-type motions. A t that time, the major mathematical concern was whether or not this generalization was applicable for de- caying motion (negative real part of s). H This skepticism pre- vailed despite the arguments from analytic continuation. 9 Thus, work on time domain unsteady aerodynamics was at a stand- still. Some 25 years later, interest was renewed in time-domain methods. Hassig 10 used rational functions in the L aplace do- main. Vepa 11 introduced the method of Fade approximants to give a finite state representation of any aerodynamic fre- quency-domain lift f unction, 1 1 as did Dowell. 12 The authors of R efs. 13 and 14 attacked the problem for two-dimensional compressible and incompressible flows. Their work is in the L aplace domain and builds on the work of Sears. 5 It utilizes numerical methods to locate the poles and to perform the inverse L aplace transform by contour integration. 13 - 14 Work f urther developed for the three-dimensional case (finite wing), 15 - 16 but that is the subject of Part I I , the sequel to this article. I n rotorcraft aeroelasticity, the development of time-do- main (rather than frequency-domain) unsteady aerody- namics is particularly crucial due to the existence of periodic coefficients and nonlinear stall, which preclude superposi- tion of F ourier or L aplace solutions. The fundamental fre- quency-domain result was derived by L oewy. 17 I t is an ex- tension of the Theodorsen theory and assumes layers of vorticity below the airfoil to account for the returning wake. However, its application is limited to linear problems of hover and climb. Dinyavari and F riedmann 18 used F ade approximants of both Theodorsen and L oewy functions in order to accommodate some unsteady aerodynamics into periodic-coefficient F loquet stability analysis. Most dy- namic stall models for rotorcraft (e.g., R ef. 19) utilize the Wagner function in a convolution integral to account for the time variation in induced flow due to the vorticity shed from stall. O ne exception is the O N ER A dynamic stall model, 20 in which a first-order differential equation pro- vides a single-state approximation to the Theodorsen func- tion. (This is in contrast to the normal two-state approxi- mations, R efs. 8 and 18.) I n R ef. 21, the O N ER A model is generalized as a vorticity-based model, but still with a one-state Theodorsen model. I n R ef. 22, the one-state model is replaced by a hierarchical f inite state inflow model for rotors. Present Approach In this article we offer a new type of finite state aerodynamic model. The model offers finite state equations for the induced flowfield itself. These equations are derived directly from the potential flow equations (either velocity or acceleration po- tential). Thus, no intermediate steps are invoked in which restrictions are placed on blade motions; and the theory is an arbitrary-motion theory from the outset. I n contrast to CF D and vortex lattice methods, the states represent induced flow expansion fields rather than velocities at discrete nodes. A s a result, the states are hierarchical, and the equation coeffi- cients are known in closed form. N o numerical fitting of fre- quency-response or indicial functions is needed. F urthermore, the induced-flow expansion implies that only a few states are needed, and the number of states can be chosen a priori based either on the texture required in the induced flowfield or on the frequency range of interest. The resultant equations are easily coupled with structural or con- trol equations and can be exercised in the frequency-domain, L aplace domain, or time domain. The above approach can be followed either in two- or three- dimensional flow. I n this article, Part I , we consider two- dimensional flow about a finite strip (i.e., airfoil) and apply the nonpenetration condition to recover thin airfoil theory. I n the sequel, Part I I , we will apply the approach to the three- dimensional flow about a disk (i.e., rotor). The disk is taken as penetrable (an actuator disc) for application to rotorcraft rather than circular wings. F or flow near the blades, Part I I utilizes the lift equation of Part I , but with the three-dimen- sional induced-flow model. Thus, this article forms a basis for rotor inflow as well as for two-dimensional aerodynamics. Theoretical Background Fluid Mechanics The airfoil lies on the segment y = 0, -1 < x < +1 in the nondimensional coordinate system of A ppendix A . If we define u and v as the nondimensional x and y velocity com- PETER S, KA R U N A MO O R T HY, A N D CA O : T WO -DI MEN SI O N A L THI N A I R F O I L 315 ponents, and if we assume the freestream velocity (normalized to unity) is in the f direction, then the continuity and mo- mentum equations are du dv_ _ dx dy du du d<& _ i _ ____ dt d ~ dX (with zero initial conditions). I t follows that dv_ dv_ _ -d< dt df dy We then define special potential functions <& v and d<$ A -du d$> A -dv (1) dx dx dt - du dy dy dt -dv (2) The continuity and momentum equations are then fulfilled if <& A and <& v satisfy The velocity field may be found from <& v by dx df (4) dy ^ The only pressure discontinuity allowed is across the airfoil, and specification of the time history of this discontinuity de- fines the pressures and velocities everywhere. For the case of edgewise flow (f = x), we have = I J -dx df, - / : dt d f v - dt J- djc - IT* (5) (6) It follows that there is a strip of concentrated vorticity y that exists on y = 0, - 1 < x < o> . O n the airfoil ( - ! <*<+! ) it is the bound vorticity y/,; and behind the airfoil (1 < x < 30) it is the wake vorticity y, v . From the definition of vorticity and the integration around a small loop across this strip, one can relate y, <J > K , and P D (the pressure across the vorticity sheet): y =<S>v - <&V, P/)=QL_4,^ (7) From Eq. (6b) we then have the pressure-vorticity relation = 7,, . at J - i f 0 = y n . + f + - J y n . d * = j i y h dx = - J +| y n . (8a) (8b) (9) dP D dx dy h dx dy h dt . . , w 0 =-^+-^ dx dt y H ,(*,0 = - f(f -x +1) (lO a) (lO b) (lO c) Clearly, the velocity u is discontinuous across the vortex sheet; but the normal velocity v is continuous. Thus, <J > V is discon- tinuous, but d<& v / dx is continuous. Equations for Bound Vorticity In order to obtain a set of state-variable equations for the induced flow, we can treat the circulation-based equations in the previous section with an expansion. To begin, we consider the induced flow due only to y h . This induced flow is inde- pendent of the time history of y h . Thus, due to the asymmetry about y = 0, we can write $> v (and y/,) as expansions in the potential functions of A ppendix A where j takes on the values s, a, 1, 2, 3, 4, . . . , and A implies evaluation on the upper surface. The velocity field from this bound vorticity then follows either from the Biot-Savart law, 23 or from Eq. (4b): -v = downwash = (12) Thus, the y, represents both bound vorticity coefficients and downwash coefficients, which are exactly the Glauert velocity expansions. R eturning to the dynamic case, we see that these downwash coefficients can be related to the airfoil differential pressure by Eq. (8a). Thus, if we expand the pressure in a series similar to that for vorticity then we obtain from Eq. (8a) (14) However, from A ppendix A we see that the ^ can be uniquely expressed in terms of the <f> y . This allows a balancing of coef- ficients in Eq. (14), since the <f> y are linearly independent. The result of this balancing is a set of ordinary differential equa- tions that relate the velocity coefficients y y and the pressure coefficients T,. % = T V, y a = r a (I 5a) y a ~ ^ 2 +r, = r, - 2f (isb) (\l2n}(y n ^ - y a +l ) +y n = r n - (2ln)T n = 2, 3, 4, . . . (15c) where f is the normalized total bound vorticity f ^ I727T = y s +h, (16) Wake Vorticity The next step in the derivation is to find equations for the induced flow due to. shed wake vorticity A . From R ef. 23, we 316 PETER S, KA R U N A MO O R T HY, A N D CA O : TWO -DI MEN SI O N A L THI N A I R F O I L see that this component of downwash can be expanded (on the airfoil) in a similar fashion as w, Eq. (12) A = y (17) where A ,, = A ( ) of R ef. 21 and A s is not used since ^.' = 0 on - 1 < x < +1 . R eference 23 provides formulas for A y in terms of the wake vorticity. F or simplicity, we write these as func- tionals: ~ 77 J 1 = -- f y lv g(77) 77 Jo dry f em r e - " " i r i =Q r~T7^=Q Lsmh(r / )J L " We note that the above formulas imply that a differential equation for this functional can be obtained based on Eq. (lO c) and integration by parts dc am =- = 2fg(0) dg/dr? 1 (19) When Eq. (19) is applied to the A , functional in Eq. (18), one must insure that g(0) = /(I ) is finite. Therefore, we use A ,,. , - A / } 4 1 as the left side of Eq. (8), which gives a #(17) of the form: g(ri) = [ e ~ ( n 1)T J - e- ( /l +1)7J ]/sinh r/ = 2e~ fjr > (20) The resultant differential equations are A ,, - U 2 +A , = 2f (l/ 2n)(X n _ l - A , I +I ) +A /f = (2ln)T n = 2, 3, 4 . . . (21) The similarity with the y- } equations [Eqs. (15)], is striking. Boundary Conditions It is especially interesting to use Eqs. (15) and (21) to form differential equations for the total induced flow w (downwash due to bound plus wake vorticity). If we expand w as (22) J ^ s i Wj = y, + A ,, / s then, by addition of Eq. (15) and (21), we have It is important to note in Eq. (23) that w (the total down- wash) is completely determined by the nonpenetration bound- ary condition. Thus, for small airfoil deformations y(x, t ) , -1 < x < +1 dy dy + dx dt (24) Therefore, the only unknown in the entire airfoil loading (r y ) is A ,,. O f particular interest are the normalized lift L , the normalized pitching moment about midchord M, and the mo- ment about the quarterchord M 1/4 : +ir, M = -k - ir M 1/4 =M - i (25) From Eqs. (23) and the Kutta condition (r v = r a = y s y,,), we then can write L = w a + ? Wi - \ a + i(vi> ,, M = i(iv fl - \ a ) - iw 2 ~ ^ ( (26a) (26b) ( 26c ) Historically, w (l + iw, has been called the quasisteady lift, w + ? w i ~ ^ tne circulatory lift, and i(vv rt - ivv 2 ) the noncirculatory lift. Since M 1/4 (and all higher pressure coef- ficients r ;i ) are completely determined by airfoil motion, we concentrate on L and A ,, for the remainder of this article. The power of Eq. (26a) for L in terms of A rt is that the A A 7 solution does not need to have the fidelity necessary to meet explicit boundary conditions. We need only an accurate A ,,. Equation (21) gives relationships for the A ,,. U nfortunately, A ,, cannot be determined from Eq. (21) alone. O ne additional relation is needed that relates A ,, to the A ,,. From the operator notation in Eq. (18) and the shorthand notation Z = e~^, we see that if one could approximate unity on 0 < 17 < o by b n Z" 0 < Z < 1 then one could approximate A w by iN A ,, (27) (28) A lthough Eq. (27) can never hold at Z = 0, the approximation must converge only on the open interval (excluding Z = 0 at 17 oo). Thus, unity can be represented as a power series. Closed-form results for the b n are given in A ppendix C. The result is a set of finite state equations for the A /7 and (more importantly) for A ,,. The equations are comprised of Eqs. (21) and (28) with = y a 2-7, = (29) These, along with Eqs. (26), form a finite state theory of unsteady aerodynamics in which the states are the Glauert inflow distributions. (\Kn)(* n _ i -w n+l ) + w n = r n (23) These equations can also be obtained from a direct application of the acceleration potential, A ppendix B and R ef. 24. Application Matrix Form I n order to verify the finite state aerodynamic model given by Eqs. (21), (26a-26c), (28), and (29), we apply it to the cases of the Theodorsen and Wagner functions. To facilitate this application, the theory is put into matrix form. F irst, we PETER S, KA R U N A MO O R T HY, A N D CA O : TWO -DI MEN SI O N A L THI N A I R F O I L 317 define the following matrices (in addition to the expansion coefficients , A ppendix C): D = \ n (n = m +1) D nm = -n (n = m - 1) D I I I H = 0 n + m 1 (30) 4, = i n = 1 d n = 0 A Z ^ 1 c n = 2/n (31) Then we rewrite Eq. (21) as [D + db r }{\} + {A} = {c}f (32) where f = (w (l +iw,) - (A ,, +U ,) L = w (l + 3H> , - A ,, +i(n> ,, - 3H> 2 ) (33) A lternately, we may write [A]{\] + {A} = {c}(* tl + iiv, ) (34) where [/I] - [D +d/? r +c d T +id> r ] (35) Special Cases The Theodorsen function is defined as the ratio of circu- latory lift to quasisteady lift (in the complex domain) for motion of the form e ikt . Thus C(k) = ^ w (l + A w, T + A ,, + ? ) For A ,, = \ n e ikt , f= e ikl , we have (36) {A} = [(D + db T )ik + I]~ l {c}ik A ,, - 3^ A ,,, U , = d r A (37) Equation (37) defines the Theodorsen function for any N. I t is interesting to contrast the simplicity of form of Eq. (37) with the complexity of the Bessel functions in Theodorsen theory. The Wagner function is defined as the circulatory lift w (l + I H> , - A ,,, for the special case w tl = u( t ) , step function, and H> ! = 0. From Eq. (26), this reduces to W(t) = I - A ,,(0 = ' l - (38) where A ,,(0 is the solution to Eqs. (34), i.e., [A}{\} + {A} = {c}8(0, { A (0)} = (0) (39) or [A}{\} + {A} = {0}, { A (0)| = [A]-*{c] (40) Equation (40) can be solved by time-marching or by utilization of the eigenvalues and eigenvectors of [A], T n , and B . (T n are time constants of the flow.) [B ]-*[A][B ] = [T n ] { A (0} = [B ][e-' = (B ][e-" T / T a ][B ]-*{c} (41) The limiting cases of the Theodorsen and Wagner functions are determined by the b n . A n interesting limit is the slope of the imaginary part of C(k) as k approaches zero. Theoreti- cally, as k approaches zero the slope of C(k) goes to infinity like ///(/:). I nterestingly, one can show that the finite state model gives this slope as lim d(ik) (42) A ppendix C shows that this sum is an approximation to ///(O ), and approaches o as A ^ increases. A nother interesting limiting case is the limit of C(k) as k approaches o, and the limit of W(t) as t approaches zero. T heoretically, these should be equal to each other and equal to 0.5. A lthough the algebra is a little involved, one can show that: 1) if 2 b n = I and 2 nb n = 0 (as in the binomial ex- pansion for b n , A ppendix C), then C(k) and W(t) will equal exactly 0.5 in the finite state model; and 2) if 2 b n = 1, then C(k) and W(t) will approach 0.5 as A ^ becomes large. For example, for W odd lim C(k) = l\ = 2, nb n / ( i, = 1.2,3. / L + 1) (43) Since the 6,, alternate signs as (-1)" +1 , W(Q) begins at 0.6 f o r A ^ = l,bi = 1, and quickly approaches 0.5 as Wincreases. L astly, one can show that the sum of all the eigenvalues of [A] equals the trace [A]. Thus, T n = 1 k bjn (44) which also approaches ^ as N becomes large. Results We would now like to examine the convergence of the finite state model for various choices of b n . F igure 1 compares the R e[C(/c)], I m[C( )], and W(t) with the finite state approxi- mations for N = 4. In order to see the performance over the entire range of k or r, the Theodorsen function is plotted vs kl(l + k), and the Wagner function is plotted vs r/(2? r + t). Exact solutions are obtained numerically. O ne can see that the augmented least-squares series for b n (see A ppendix C for closed-form) gives a reasonable fit. The other least-squares methods (not shown in the figure) give a similar result. The binomial expansion, which is designed to give least error at large k (and small /) , does so, but to the detriment of small k (or large t) response. F igure 2 shows a similar comparison for N = 8. The error is greatly reduced, although the binomial expansion is converging slowly. F igure 3 presents a quantitative measure of an error norm defined as the rms error computed on 100 equally spaced points on the k/ (l + k) or t/ (27T + r) plot. The error is normalized such that the N = 0 case, C(k) = W(t) = 1, has an error of unity. We see that the binomial expansion con- verges at a rate of N~ l , whereas the augmented least squares converges as N~ 2 . However, due to the large factorials in the augmented method, double precision arithmetic fails for W > 10 for C(k) and N > 8 for W(t). The minimum error for the augmented least squares occurs at N = 8, and is 1% for C(k) 318 PETERS, KARUNAMOORTHY, AND CAO: TWO-DIMENSIONAL THIN AIRFOIL 0.8 -- ReCOOOJ - 0.6 -- 0.4 Hh 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.3 -r -ImC(k) 0. 2 0. 3 0. 4 0. 5 0. 6 0. 7 0. 8 0. 9 1 C) T/ (T+2jt) Fig. 1 a) Real part of C(k}, N = 4; b) imaginary part of C(k}, N = 4; and c) Wagner function, N = 4. 0. 1 0. 2 0. 3 0. 4 0. 5 0. 6 0. 7 0. 8 0. 9 1 0. 2 0. 3 0. 4 0. 5 0. 6 0. 7 0. 8 0. 9 1 C) T/(T+2lC) Fig. 2 a) Real part of C(k), N = 8; b) imaginary part of C(k), N = 8; and c) Wagner function, N = 8. and 1.3% for W(t). The binomial expansion (with smaller factorials than the augmented least squares) begins to diverge after N = 50 for C(k) and TV = 16 for W(t). (R ecall that eigenvalue problems are more sensitive than are matrix in- versions.) The minimum error is 0.7% for C(k) (N = 50) and 5% for W(t) (N = 16). Quadruple precision would extend the accuracy of either method. Table 1 provides a summary of these errors and compares them with those from J ones' approximation (which has two states), and with Fade approximants with two and three states, respectively. N ote that the J ones' method, which is based on a fit of the Wagner function, has a lower error norm for W(t) than it does for C(/c). Conversely, the Fade results, based on a least-squares fit of C(/c), have a lower error norm for C(/c) than they do for W(t). It is at first puzzling to see that one needs four to six inflow states with the augmented method to obtain the same accuracies as can be obtained with only two to three states for J ones' and F ade methods. The reason for PETERS, KARUNAMOORTHY, AND CAO: TWO-DIMENSIONAL THIN AIRFOIL 319 Table 1 Error norms Methods Theodorsen 0.1 -- ERR 0.01 -- 0.001 a) ERR 0.01 -- 0.001 b) Wagner Curve-fit J ones F ade, N = 2 Fade, N = 3 A ugmented N = 4 N = 6 N = 8 Binomial N = 4 N = 8 N = 16 N = 50 2.3% 2.1% 1.2% 3.5% 1.6% 1.0% 10.5% 5.6% 2.8% 1.3% 0.7% 1.8% 3.8% 3.4% 5.0% 1.8% 1.3% 18.3% 10.1% 5.0% __________ Fig. 3 Error norm for a) V(Re C) 2 + (Im C) 2 and b) this phenomenon is found in a study of the eigenvalues of the system. The accuracy of the finite state model is obtained almost entirely from the first two to three eigenvalues (which are real), even for N = 50. Thus, only a few states are re- quired; but they are a combination of many of the Glauert states. O ne might suspect, therefore, that a change of variable could improve the convergence of the finite state results. Conclusions A finite state induced flow model has been developed from first principles (as opposed to numerical fits of Theodorsen or Wagner functions). The method is hierarchical, and the states represent the classical Glauert induced flow coeffi- cients. The resultant, closed-form equations for the inflow and states give excellent correlation with Theodorsen and Wagner functions with four to nine states included. This finite state type of aerodynamic analysis is very useful in aeroelas- ticity because it can be used in the frequency domain, L aplace domain, or the time domain. F urthermore, the method is more rigorous than other finite state methods in that it is based on first-principle fluid mechanics rather than on curve- fitting specialized response functions; and the method is more computationally direct than methods based on extensions in the complex plain that require extensive numerical analysis. The weakness of the method is that convergence is not as rapid as is theoretically possible. A n appropriate change of variable in the inflow expansion could cure this. Appendix A: Potential Functions with Elliptical Coordinates Coordinates We define an x, y coordinate system with positive x down- stream and positive y in the direction of positive lift. A ll lengths are nondimensional on 6, such that the airfoil is on the line segment l < ^ < +l , _y = 0. Pressures are nor- malized on pV 2 , and time is normalized on the semichord over velocity. (It follows that velocities are normalized on V.) The transformation from elliptical coordinates to Cartesian coordinates is given by x = cosh 77 cos </> , 0 < </> < 277, _y = sinh 77 sin </> 1 <77<0 0 (A l) Thus, upstream </> = TT, x = -cosh 17, y = 0; downstream <p = 0, x = cosh 77, y = 0; and on the airfoil 77 = 0, ;c = cos </> , y = 0. The transformation of derivatives is given by dx dy where J _ sinh 77 cos </> cosh 77 sin cf> A ^ cosh 77 sin (/> sinh 77 cos < (A 2) A = sinh 2 77 + sin 2 </> = cosh 2 77 - cos 2 ^ = sinh 2 ? ] cos 2 (/> + cosh 2 r/ sin 2 (/> Laplace's Equation L aplace's equation is ^ + ^ = ^ + T 7 l ) = 0 < A 3 > Through separation of variables, we obtain the fundamental solutions 1, (/>, 77, 77(/> e' lri When these solutions are used to represent pressure, the pres- sure must vanish at infinity and have discontinuities only on the airfoil. Thus, acceptable pressure functions are < = e flri sin(ft</> ) n = 1, 2, 3, . . . (A 4) Due to the A 2 in the denominator of Eq. (A 3), we must also 320 PETERS, KARUNAMOORTHY, AND CAO: TWO-DIMENSIONAL THIN AIRFOIL consider derived potential functions that have singularities at A = 0 (leading and trailing edge of airfoil). We define sym- metric and antisymmetric functions If we expand this concept, we can develop differential equa- tions for Wj based on Eq. (3a): = X 2 ^ // = t sin 0 cosn n= 1,3,5 (series converges 17 =t= 0) = 2j (~^) ( * > n = - [sin 0 cos //= 2,4 ,C> (series converges 77 =1= 0) (A 5) We also define < , = 4> t - 4 < v , which yields zero net pressure across the airfoil. It follows from Eqs. (4) and (5) that w y = r/, and that (B2) The nature of the acceleration potential method is that the pressure (or velocity) expansions should balance on the air- foil. The <J > ' , ^ ' relationships in A ppendix A show that Eq. (B2) is balanced when Velocity Potentials We also define a set of velocity potentials based on (A 6) TT <f> = X 2^ ,, (series converges everywhere) /i = l , 3, 5 - e" 7 7 sin </> = ~ 2,, / i=2, 4, 6 (series converges everywhere) 2 \n + 1 n - 1 (n = 2, 3, 4, . O n the airfoil (TJ = 0) 0) (A 7) (A8) Gradients of Potentials L et ( )' = d( )/ dy. This gives <$> ,', as the normal acceler- ation and W n the normal velocity: <E> S ' = sinh 17 cosh ^ [(cosh 2 ? / + cos 2 </> )sin 2 c/> -f (sin 2 < sinh 2 r/)cos 2 (/)]/A 3 <& ', = sinh r\ cos </> [sinh 2 i7 cos 2 ^ - 3 cosh 2 r/ sin 2 </)]/A 3 O,', = He~ l l l 7 [cosh 17 sin c/> sin(i7(/> ) - sinh r/ cos </> cos(i7^ > )]/A (A 9) ^ ' = sinh 17 cos $/A _i I ,, = T H = 2, 3, . . . (B3) This is identical to the result from our thin-airfoil derivation, Eq. (23). However, the above equations are not complete since they omit the distributions w s and w a . The singularities associated with the derived functions 4 > v , <!> , W s imply that the theory must be extended to include these singularities in order to obtain w s , w (l equations. For the special case of no- shed vorticity (F = y v + iyi = 0- w x + w { = 0, A y = 0), the W' s terms disappear; and w tl = r (n vv s = r, become the two extra equations. These cause the velocity balance to be fulfilled everywhere in the flowfield. These, then, agree with thin-airfoil theory (y, = vv y ), Eq. (15a). F or the general case, the equation cannot be balanced everywhere; and the far field must become a "natural"bound- ary condition. The balance of l/sinh 3 ry leading-edge singular- ities gives (T V w> v ) = (T U (B4) I n other words, the classical thin-airfoil pressure distribu- tion gives the classical inflow. (This is also equivalent to A ,, = -A s , which implies no leading-edge singularity due to shed vorticity.) The balance of 1/sinh 17 singularities gives a second equation 2Z - (w, + w rt ) = 0 (B5) where Z ^ ^ ~ T J . A t 17 = 0, Z = 1, this gives w a ) = 0 (B6) a divergent (although functionally correct) series. To manip- ulate this into a better-behaved series, we divide by Z and integrate from 0 to 1. The result is 1 +Z - o (B7) Appendix B: Acceleration Potential In Eq. (11), <& v is expanded in a <f> y series for the static case, in order to find the induced flow due to bound vorticity. where ///(()) is approximated by 2 b,J n (A ppendix C). This series still fails to converge at Z = 1, (1 = 2 - 2 + 2-2 + 2 - 2 . . .). However, when the diverging series of 2 is PETER S, KA R U N A MO O R T HY, A N D CA O : TWO -DI MEN SI O N A L THIN A I R F O I L 321 replaced by a series of b n (A ppendix C), we have a well- behaved model: (T,, - w a ) = \^b \(r n - nO - - K + *) (B8) 2 \_ n j A s a special case, consider no inflow due to bound vorticity: Ta = 0, T,, = (2//i)T
, Eqs. (15). The term w
s + w a = T S + r a = r v cancels every b n r n term yielding b n w n (B9) This is the same as from the vorticity equations, Eq. (28). A lso note that the w s + w a coefficient, 2 b n / n, approaches 30, -///(O ), as N becomes large. Appendix C: Shed-Wake Expansion Consider the approximate expansion ^ z "' o < z < i If we divide by Z and integrate from 0 to 1, we have Evaluation at Z 1 yields the following: (ci) (C2) (C3) Thus, 2 bjn will approach oo as A ^ is increased. To apply a weighted least-squares calculation of b n , we minimize /, where J = dz (C4) Y(Z) is a weighting function. Different weights will result in different inflow theories. For Y(Z) = Z* (Z = 1 is the most important point), we obtain an expansion this further implies b n = (-1)"- - 1 - (1 - Z) A N\(N - n)\ ' (C5) nb n = 0 (C6) These are the binomial expansion coefficients of (1 - Z) N . For Y(Z) = Z (/ , q = integer, we have a family of b n . The case q = - 1 (smallest possible integer value) gives a closed- form expression b = (-p.- ^ ; (C7) In this formulation, the 1, b n = 2 for TV odd, and ^ b n = 0 for W even. From Eq. (18) we see that the functional form of the A ,,, A ,, relationship is 1 (C8) sinh r] /I = i sinh 17 This shows that the error in the r\ functional involves a sin- gularity at 77 = 0. Thus, for the least-square error from Eq. (C8) to be finite error = one must have 2 b n = 1 in order to cancel the singularity. Thus, we also consider modifications of Eq. (C7) to meet this criterion. In particular, we define three possibilities: 1) mod- ified least squares, b N replaced by b N + (- 1) N ; 2) augmented least squares, b n taken from formula for N - 1 terms and b N = (- l) N+1 ; and 3) least squares fit (q = - 1) with formal constraint on E b n = 1. Acknowledgments This work was sponsored by the U .S. A rmy A eroflight- dynamics Directorate (A TCO M), N A SA Grant N A G-2-728, R obert A . O rmiston, Technical Monitor. References 'Wagner, H., "U ber die Entstehung des dynamischen A uftriebs von T ragflugeln," ZA MM, Bd. 5, Heft 1, Feb. 1925, pp. 17-35. 2 Theodorsen, T., "General Theory of A erodynamic I nstability and the Mechanism of F lutter, 11 N A CA R ept. 496, May 1934, pp. 413- 433. "Garrick, I . E., k t O n Some R eciprocal R elations in the Theory of N onstationary Flows, 11 N A CA R ept. 629, 1938. 4 J ones, R . T., "O perational T reatment of the N onuniform L ift Theory to A irplane Dynamics, 11 N A CA TN 667, March 1938, pp. 347-350. 5 Sears, W. R ., "O perational Methods in the Theory of A irfoils in N on-U niform Motion, 11 J ournal of the Franklin Institute, Vol. 230, J ul y 1940, pp. 95-111. Mones, R . T., "The U nsteady L ift of a Wing of F inite A spect R atio, 11 N A C A R ept. 681, J une 1939, pp. 31-38. 7 J ones, W. P., "A erodynamic Forces on Wings in N on-U niform Motion, 11 British A eronautical R esearch Council, R & M 2117, A ug. 1945. <s Bisplinghoff, R . L ., A shley, H., and Hal fman, R . L ., Aeroelas- ticity, A ddison-Wesley, R eading, MA , 1955. ''Edwards, J . W., Breakwell, J . V., and Bryson, A . E., J r. , "R eply by A uthors to Vepa, 11 J ournal of Guidance and Control, Vol. 2, N o. 5, 1989, pp. 447, 448. '"Hassig, H. J . , "A n A pproximate True Damping Solution of the F l utter Equations by Determinant I teration, 11 J ournal of Aircraft, Vol. 8, N o. 11, 1971, pp. 825-889. "Vepa, R ., "O n the U se of F ade A pproximants to R epresent U nsteady A erodynamic L oads for A rbitrarily Small Motions of Wings, 11 A I A A Paper 76-17, J an. 1976. l 2 Dowell, E. H., "A Simple Method for C onverting F requency Domain A erodynamics to the Time Domain, 11 N A SA TM-81844, O ct. 1980. "Edwards, J . W., A shley, H., and Breakwell, J . V., "U nsteady A erodynamic Modeling for A rbitrary Motions, 11 A I A A Paper 77- 451, March 1977. l 4 Edwards, J . W., Breakwell, J . V., and Bryson, A . E., J r. , "A c- tive F lutter Control U sing Generalized U nsteady A erodynamic The- ory, 11 J ournal of Guidance and Control, Vol. 1, J an.-F eb. 1978, pp. 32-40. 15 Matsushita, H., "A erodynamic Transfer F unctions of a F inite Wing in I ncompressible F low, 11 20th A ircraft Symposium, F ukuoka, J apan, N ov. 1982. l 6 Miyazawa, Y., and Washizu, K., "A F inite State A erodynamic Model for a L if ting Surface in I ncompressible Flow, 11 AIAA J ournal, Vol. 21, N o. 2, 1983, pp. 163-171. l 7 L oewy, R . G., "A T wo-Dimensional A pproximation to U nsteady A erodynamics in R otary Wings, 11 J ournal of the Aeronautical Sci- ences, Vol. 24, N o. 2, 1957, pp. 81-92. l s Dinyavari, M. A . H., and F riedmann, P. P., "U nsteady A ero- dynamics in Time and F requency Domains for F inite Time A rbitrary Motion of R otary Wings in Hover and F orward F light, 11 A I A A Paper 84-0988, May 1984. 1M Beddoes, T. S., "R epresentation of A irfoil Behavior, 11 Vertica, Vol. 7, N o. 2, 1983, pp. 183-197. 2() T ran, C. T., and Petot, D., "Semi-Empirical Model for Dynamic 322 PETER S, KA R U N A MO O R T HY, A N D CA O : T WO -DI MEN SI O N A L T HI N A I R F O I L Stall of A irfoils in View of the A pplication to the Calculation of R esponse of a Helicopter Blade in F orward F light," Sixth European R otorcraft and Powered L ift F orum, Bristol, England, U K, Sept. 1980. 2l Peters, D. A ., 'Toward a U nified L ift Model for U se in R otor Blade Stability A nalyses,' 1 J ournal of the American Helicopter So- ciety, Vol. 30, N o. 3, 1985, pp. 32-42. -Peters, D. A ., and Su, A ., "A n I ntegrated A irloads-I nflow Model for U se in R otor A eroelasticity and Control A nalysis, 1 ' Mathematical and Computer Modelling, Vol. 19, N os. 3/4, 1994, pp. 109-123. 23 Dowell, E. H., Curtiss, H. C., J r. , Scanlan, R . H., and Sisto, F ., A Modern Course in Aeroelasticity, Sij thoff and N oordhoff, A l- phen aan den R i j n , The N etherl ands, 1980. 24 Cicala, P., "L e A zioni A erodinamiche sui Profili di ala O scillanti in Prezenza di Corrente U niforme, 11 Memoirs of the R oyal A cademy of Science, Torino, Serie 2, Tomo 68, Parte I, J uly 7, 1935. Ro t ary Wing S tructural D y namics and A ero elas t ic it y Richard L Bielawa This new text presents a comprehensive account of the fundamental concepts of structural dynamics and aeroelasticity for conventional rotary wing aircraft as well as for the newly emerging tilt-rotor and tilt-wing concepts. Intended for use in graduate level courses and by prac- ticing engineers, the volume covers all of the important topics needed for the complete understanding of rotorcraft structural dynamics and aeroelasticity, including: basic analysis tools, rotating beams, gyroscopic phenomena, drive system dynamics, fuselage vibrations, methods for Place y o u ro rd ert o d ay ! Call 1-800/682-A IA A &ALAA A meri c an In s t it u t e o f A ero n au t i c s an d A s t ro n au t i c s Publications Custome r Se r vice , 9 Jay Could Ct. , P. O. Box 753, Waldor f, MD 20604 FAX 301/ 843- 0159 Phone 1 - 800/ 682- 2422 8 a. m. - 5 p. m. Easte r n controlling vibrations, dynamic test procedures, stabil- ity analysis, mechanical and aeromechanical instabili- ties of rotors and rotor-pylon assemblies, unsteady aerodynamics and flutter of rotors, and model testing. The text is further enhanced by the inclusion of prob- lems in each chapter. AIAA Education Series 1992, 584pp, illus, IS BN 1-56347-031-4 A IA A Members $54.95 No nmembers $75.95 Ord er*: 31-4(830) Sale s Tax: CA r e side nts, 8. 25%;DC, 6%. For shipping and handling add $4. 75 for 1 - 4 book s (call for r ate s for hig he r quantitie s). Or de r s unde r $100. 00 must be pr e paid. For e ig n or de r s must be pr e paid and include a $20. 00 postal sur char g e . Ple ase allow 4 w e e k s for de live r y. Pr ice s ar e subje ct to chang e w ithout notice . R e tur ns will be acce pte d w ithin 30 days. l\lon- U. S. r e side nts ar e r e sponsible for payme nt of any taxe s r e q uir e d by the ir g ove r nme nt.