Вы находитесь на странице: 1из 9

Journal of Sound and Vibration (1989) 131(2), 265-273

STABILITY BOUNDARIES

FOR PARAMETRICALLY EXCITED

SYSTEMS BY DYNAMIC STIFFNESS


A. Y. T. LEUNG
Department of Civil and Structural Engineering, University of Hong Kong, Hong Kong

(Received 19 July 1988, and in revised form 3 January

1989)

The dynamic stability of skeletal systems subject to harmonic axial forces is of interest. Temporal discretization is achieved by Fourier expansion. The resulting differential equations in spatial co-ordinates alone are solved by the exact frequency-dependent shape functions. The dynamic stability boundaries are determined by studying the free vibration behaviour with periods T and 2T, where T is the period of the harmonic axial force. Since spatial discretization is completely eliminated, many stability boundaries can be determined accurately with the minimum number of elements.
1. INTRODUCTION

shape functions, which are stiffness matrix is formed by frequency-dependent exact solutions of the governing differential equations. It eliminates spatial discretization errors and predicts the infinite number of vibration modes accurately. The method has been applied with success to many dynamic problems including natural vibration [l- 121 and response analysis [13,14]. It is extended here to the determination of the dynamic stability boundaries of skeletal structure under parametric excitation. The dynamic stability of elastic systems was discussed extensively in the text by Bolotin [15], where the solution methods were confined to modal analysis. While the Fourier expansion is effective for time discretization, the finite element method is also successful for spatial discretization [16,17]. The spatial discretization is alternatively achieved by frequency-dependent shape functions, which are the exact solutions of each element under mono-frequency vibration. Since the shape functions vary according to vibration frequency, the method is exact for mono-frequency vibration and is able to predict the infinite number of vibration modes with very few number of degrees of freedom. However, even if the parametric excitation is mono-frequency, the Fourier transformation results in a set of coupled ordinary differential equations in the spatial variable with various frequencies of interest. A mixed frequency formulation is introduced here to obtain the dynamic stiffness matrix, the determinant equation of which has roots corresponding to the stability boundaries. The Galerkin approximation is effectively introduced. A convergence study is carried out, by taking more elements than necessary. Since the mono-frequency shape functions are very close to the exact solutions for multiple frequency vibrations, the frequency components of which are clearly distinct in the case of parametric excitation, it is found that the calculated stability boundaries are not affected by the number of elements being used in the numerical examples.
A dynamic 2. FORMULATION Consider a beam member, equation for lateral vibration isolated from the system. The governing partial differential u(x, T) under parametric excitation of period T = 27r/ 8 is:

EI a%/ax4+ (P + 24 cos 7) a*v/ax* + me* C?u/c%*= 0,


265
002240X/89/ 140265+ 09 $03.00/O

(1)

@ 1989 Academic Press Limited

266

A. Y. T. LEUNG

where EI is the flexural rigidity, P + 2Q cos 7 is the time-varying axial force, T = 8t is the non-dimensional time, and m is the mass per unit length. Since the set of trigonometric functions is complete, for steady state periodic vibration, the following Fourier expansion in terms of the unknown functions C-Q(X) and bk(x) is admissible: V(X,7) =&(x)+x
k

[uk(x) sin (k7/2)+bk(x)

cos (k7/2)].

(2)

Here the summation over k is from one to n, the number of terms under consideration. The dynamic stability boundaries are determined by the non-trivial solutions of &(x) and bk(x) having periods T and 2T, respectively. Non-trivial solutions with period T are given by equation (2) when k is even, and those with period T when k is odd. Substituting equation (2) into equation (l), multiplying by sin (jT/2) and cos (jT/2) respectively and integrating 7 from 0 to 2m, one has El(d4/dx4){a}+[PI+ QJ](d*/dx*){a}Qa,{e,]-[M~]{aO} = (01, (3) (4) (5) (6)

EI(d4/dx4){b}+[PI+QJ](d2/dx2){b}+Qb,{e,}-[M~]{b}={O}, EI(d4/dx4){a }+ [ PI+ QJ](d*/dx*){a } - [MZ]{a = {O}, }

EI(d4/dx4){b }+[PI+QJ](d2/dx2){b }-[M;]{b }={O}, where {a? = 1a1,a3,a5,...

IT,

{bO}=[b,,bj,bs,...l=,
M = III, (4%. . .I , D = [WI, GI
[Mi] = me* diag [0, l*, 2*, . . .],

WI = [a29a49a6y.,.

IT> WI = @o, h, b,, be,. . .I=,

[I] = diag [l, 1, . . .]

[Mz] = (m0*/4) diag [I*, 3*, 5*, . . .],

[Mz] = me* diag [l*, 2*, 3*, . . .],


and

[Jl =

:I 1
0 1

0 1 1 101

0 .

It is difficult to solve the set of ordinary differential equations (3)-(6). Since the primary vibration frequencies for the unknowns ok(x) and bk(x) are respectively k0/2, which will not be coincident or close to each other for different S it is reasonable to express the unknowns as
akb) = [Nk(R x)I{Pk)

and

bk(x) = [Nk(R,

x)hk),

(7)

where [Nk(R, x)] is a four-element solution for the vibrating beam, EI(d4/dx4)4

shape function matrix. Each of .the elements is a + R(d*/dx )4 - m(k0/2)*+ = 0, (8)

with the following four sets of boundary conditions, respectively: (i) (ii) (iii) (iv) 4(O) = 1, 4 = 1, (O) 4(Z) = 1, 4 = 1, (I) 4(O) = cp(U = K(1) = 0, 4(O) = +(I) = K(l) =O, 4(O) = 4 = b = 0, (O) (Z) 4(O) = 4 = b(l) = 0. (O)

PARAMETRICALLY

EXCITED

SYSTEM

STABILITY

267

Here R is equal to P - Q, P + Q, or P,Iis the length of the beam concerned and a prime denotes a derivative with respect to x. The shape functions can be found, for example, from references [3] and [lo]. The nodal vectors {pk} and {qk} in equations (7) can be considered as the generalized nodal displacements for the kth Fourier component. Pre-multiplying equations (3)-(6) by {a}T, {b}T, {ae}T and {b , respectively, and } integrating over x from zero to 1, one has, respectively, W~l~p~ = WI, where [DXP = (01, I and

[D3h = WI, I
(9)

r D(f DXI =

0, e/2)
NP,

G13

G31

38/2)
G3

G,,

D(P,58/2)
.

. .I

(10)

D(P+ Q, e/2) P.%l=


G3, WP,

Gu
38/2)
G53 G,,

D( P, 58/2)

1
3

(11)

D(P,ze)

GM D(P,38) . . I

(12)

D(P,O)
[DE]=

G20 G D(~2e) e,
42 , . :1

Go2

(13)

The mixed frequency matrix [G,] is defined as

[G,l= [G(@i, = 0 1[Ni(Ri, X)lTINJ(Rjv dx, ?,)I X)1


0

(14)

where [N,(I?, x)] is given by equations (7) and primes denote derivatives with respect to x. The value of the equivalent axial force R is equal to P - Q, P + Q, or P,depending on the shape functions being used. It should be noted that [G,,], etc., in equations (10) and (11) are not the same, because in equation (lo), [NJ = [N,( P - Q,x)] and in equation (II), [NJ = [N(P+ Q, x)1. 3. SOLUTION ALGORITHM The matrices [D] and [G,] in equations (lo)-(13) are assembled element by element according to the finite element procedure. The dynamic stability boundaries are determined by the conditions det[DZ( P,Q, e)]=O, det [@(P, Q, e)]=O, det [l);(P, e)]=O, Q, where [Dz], etc., are assembled global matrices corresponding (lO)-( 13).
det

[Dz( P,Q, e)]=O,


(15)

to those given in equations

268

A. Y. T. LEUNG

Before solving equations (15), a brief account of the finite element formulated with frequency-independent shape functions is advantageous. Results, similar to equations (lo)-(13), obtained by conventional methods [ 15-171, are, respectively, K-(P-Q)G-@ M/4

-QG
K-PC-98*M/4

[fig =

-QG

-QG
K-PG-28*M/4
. I

-QG
K-(P+Q)G-8*M/4

-QG
K-PG-98*M/4

[@] =

-QG

-QG
K-PC-250*M/4

-QG
K-PC-0*M

-QG
K-PG-48*M

[ig] =
and
K-PG

-QG

-QG

-QG

K-PG-98*M

. . .I
.I

-QG
K-PG-8*M

[BE] =

-QG

-QG
K-PG-48*M

-QG

-QG

-QG K-PG-98*M
.

.
. .I

Here K, G, M are the usual stiffness, geometric and mass matrices, respectively. The relation between P, Q and 8 describing the dynamic stability boundaries is typified by det [n(P, Q, e)] = 0,
(16)

where [i)] denotes #,I, [fiz], [fi:] or [f):]. For given values of P and Q, equation (16) is reduced to a linear symmetric algebraic eigenvalue problem with A = e2 as an eigenvalue. Since [n] can be rewritten as [o(h)] =[K]-A[M], Sylvester inertia theorem [18,19] s states that for positive definite [it] and [Ml, the number of eigenvalues A which are smaller than a specific value A* is equal to the number of negative elements on the main diagonal of [b(A*)], which has been triangulated by the Gaussian elimination without interchanges. If [K] is not positive definite, then the number of negative elements on the main diagonal of the triangulated @(A*)] must be modified by the negative elements on the main diagonal of the triangulated [K] by the Gaussian method without interchanges. Therefore, by counting the numbers, one is able to locate the eigenvalues with confidence. The counting technique was generalized to frequency-dependent matrices [2] and the method is known as the Wittrick and Williams algorithm. By varying P and Q, one can compute 8 with confidence.

4. NUMERICAL EXAMPLES Consider a cantilever excited parametrically by an harmonic axial force P( 1 + 2 cos et). Using just one element for the analysis, one obtains four matrices of order 2n, where n is the number of harmonic terms considered corresponding to equations (lo)-(13). The roots of the matrix determinants for 0 as P varies give the dynamic stability boundaries. Three harmonic terms were taken and the results for equations (10) and (1 l), (12) and

PARtjMETRICALLY

EXCITED

SYSTEM

STABILITY

269

(13) were superimposed respectively in Figures l-6. Figures l(a), (b) show the approximations obtained by taking just one harmonic term and Figures 2(a), (b) that obtained by taking two harmonic terms. The co-ordinates were non-dimensionalized to p = h%%

Figure 1. Stability boundaries by one harmonic approximation:

(a) k odd; (b) k even.

4 x 3

3 P

Figure 2. As Figure 1, but two harmonic approximation.

270

A. Y. T. LEUNG

0 P

Figure 3. As Figure 1, but three harmonic

approximation.

m
Figure 4. Example frame.

PARAMETRICALLY

EXCITED

SYSTEM

STABILITY

271

x 3

Figure

5. Stability

boundaries

for the frame by one harmonic

approximation:

(a) k odd; (b) k even

(b)

Figure

6. As Figure

5, but two harmonic

approximation.

272

A. Y. T. LEUNG

and A = 4m, where PA is the mass per unit length, EI is the flexural rigidity, and I is the length of the cantilever. It is seen that fewer harmonic terms give reasonably good approximations at the onset of instability and the boundaries vary substantially when the amplitude P becomes large. More elements were taken subsequently. However, no distinction from the results shown was noted. The shaded area denotes an unstable region. The first three principal regions at the onset of instability agree with those of reference [15] obtained by using the Gale&in method. Consider the plane frame shown in Figure 4 as a second example. Five degrees of freedom per harmonic term were enough. Two harmonic terms were taken and the results for equations (10) and (ll), (12) and (13) are superimposed respectively in Figures 5 and 6. Figures 5(a), (b) show the approximations obtained by taking just one harmonic term. Similar conclusions can be drawn.

5. CONCLUSION The dynamic stiffness method is extended to parametrically excited systems. It predicts more dynamic boundaries accurately than the number of degrees of freedom employed. Numerical examples show the success of the method.

REFERENCES 1. D. C. JOHNSON and R. E. D. BISHOP 1960 The Mechanics of Vibrations. Cambridge University Press. 2. W. H. WIT~RICK and F. W. WILLIAMS 1970 Quarterly Journal of Mechanics and Applied Mathematics 24, 263-284. A general algorithm for computing natural frequencies of elastic structures. 3. V. KOLOUSEK 1973 Dynamics in Engineering Structures. London: Butterworth. 4. A. Y. T. LEUNG 1976 Ph.D. Thesis, University of Aston in Birmingham. Dynamics of structural vibrations by frequency dependent matrices and modal analysis. 5. T. H. RICHARDS and A. Y. T. LEUNG 1977 Journal of Sound and Vibration 55, 363-376. An accurate method in structural vibration analysis. 6. A. Y. T. LEUNG 1978 International Journalfor Numerical Methods in Engineering 12,1705-1716. An accurate method of dynamic condensation in structural analysis. 7. A. Y. T. LEUNG 1979 International Journalfor Numerical Methods in Engineering 14,1241-1256. An accurate method of dynamic substructuring with simplified computation. 8. J. HENRYCH 1981 The Dynamics of Arches and Frames. Amsterdam: Elsevier. 9. R. LUNDBN and B. AKESSON 1983 International Joumalfor Numerical Methods in Engineering 19, 431-449. Damped second-order Rayleigh-Timoshenko beam vibration in space: an exact complex dynamic member stiffness matrix. 10. W. P. HOWSON, J. R. BANERJEE and F. W. WILLIAMS 1983 Advances in Engineering Software 5, 137-141. Concise equations and program for exact eigensolutions of plane frames including member shear. 11. J. R. BANERJEE and F. W. WILLIAMS 1985 International Journal for Numerical Methods in Engineering 21, 2289-2302. Exact Bernoulli-Euler dynamic stiffness matrix for a range of tapered beams. 12. D. PEARSON and W. H. WITTRICK 1986 International Joumalfor Mechanical Science 28,83-96. An exact solution for the vibration of helical springs using a Bernoulli-Euler model. 13. A. Y. T. LEUNG 1987 Journal of Sound and Vibration 98, 337-347. Dynamic stiffness method for exponentially varying harmonic excitation of continuous system. 14. A. Y. T. LEUNG 1987 Dynamics and Stability of Systems 2, 125-137. Dynamic stiffness and response analysis. 15. V. V. BOLOTIN 1964 Z%e Dynamic Stability of Elastic Systems. San Francisco, California: Holden-Day. 16. B. A. H. ABBAS and J. THOMAS 1978 Journal of Sound and Vibration 60, 33-44. Dynamic stability of Timoshenko beams resting on an elastic foundation.

PARAMETRICALLY EXCITED SYSTEM STABILITY

273

17. B. A. H. ABBAS 1986 Journal ofSound and Vibration 108,25-32. Dynamic stability of a rotating Timoshenko beam with a flexible root. 18. R. A. HORN and C. A. JOHNSON 1985 Matrix Analysis. Cambridge University Press. 19. B. PARLEY 1980 The Symmetric Eigenvalue Z+oblem. Englewood Cliffs, N.J: Prentice Hall.

Вам также может понравиться