Computers & Swucrures Vol.42,No. 5,pp.853-855, 1992 Printedin GreatBritdn.

004s7949/92 $5.00 + 0.00 1992 Pcrgamon FWSS plc

TECHNICAL NOTE MATRIX FORMULATION OF THE DYNAMIC ANALYSIS OF SDOF SYSTEMS IN THE FREQUENCY DOMAIN
F. VENANCIO FILHO~ and A. M. CLARETS
TDepartment of Civil and Environmental Engineering, Rutgers, The State University of New Jersey, P.O. Box 909, Piscataway, NJ 08850909, U.S.A. SDepartment of Civil Engineering, School of Mines/UFOP, 35400 Ouro Preto MG, Brazil
(Received 14 December

1990)

Abstract-In this note a matrix formulation of the frequency domain dynamic analysis of SDOF systems is presented. The number of terms in the Discrete Fourier Transforms can be arbitrarily selected and the transforms are implicitly calculated in the procedure which leads to the response in the time domain.

INTRODUCIION of dynamic response in frequency domain is strongly indicated for structural systems with frequency-dependent properties. Linear structural dynamic analysis in frequency domain is well known and had a great development with the use of the Fast Fourier Transform (FFT) algorithm. Nevertheless, only recently methods of non-linear dynamic structural analysis in the frequency domain have been developed. Kawamoto [l] presented an iterative method called Hybrid Frequency-Time Domain method, or HFTD method, in which the non-linearities are treated as pseudo-forces. Darbre and Wolf [2] presented the segmenting version of HFTD procedure and demonstrated its convergence criterion. Venancio Filho and Claret [3] presented a method for non-linear dynamic analysis in frequency domain based on a step-by-step incremental technique with linearized steps and a secant stiffness. All these methods use the FFT algorithm for calculation of direct and inverse discrete Fourier transforms. Although the FFT algorithm is computationally very efficient it can be. very hard to work in an iterative or step-by-step non-linear analysis. The number of terms in the discrete series, using the FFT algorithm, must be a power of two. Thus, if no sufficient precision is achieved with N terms, only 2N terms can be used. This fact means that the computational effort and the storage memory needed by the algorithm can increase very rapidly. On the other hand, the FFT algorithm is apart from the procedure of the response calculation in the frequency domain, implying in the repetition of a set of operations, every time it is called. In this note a matrix formulation of the dynamic response analysis of SDOF systems in the frequency domain is presented in which the number of terms in the Discrete Fourier Transforms (DFTs) is arbitrarily selected. The only restriction is that it must be an odd The analysis

integer. An important feature of this formulation is that the DFTs are implicitly executed in the same procedure that leads to the response in the time domain. It turns out that it is very suitable for non-linear dynamic analysis in the frequency domain.
CLASSICAL FORMULATION

The dynamic response of a SDOF system in the frequency domain can be expressed by the following equations [4] o(t,) = $ and
N-l P(cii,,,) = At

i H(Gm)P(cS,) eRn(mnlw
m-0

(1)

c p(t,) e-2n(mniM.
n-0

(2)

The total time interval Tp in which the response is to be calculated is divided into N (N odd) equal time intervals given by (3) and the discrete times in which the load is defined are given by r.=nAl=n$ (Odn<N-1). (4)

The frequency range is likewise divided into N equal intervals A6 expressed as

(5)

and the discrete frequencies (3, are taken according Table Al (see Appendix 1).

853

854

Technical Note By the same token, the response from eqn (2) is written in matrix form as

In eqn (2), P(&,) is discrete Fourier transform of the load; in eqn (l), H(G,) P(&,,J is the discrete Fourier transform of the response (or the response in the frequency domain) and et(&) is the inverse discrete Fourier transform of the response (or the response in the time domain). The dynamic response expressed by eqns (1) and (2) can be numerically determined by the FFT aigorithm. In dealing with this algorithm, N must be a power of two and, consequently, an even integer. Nevertheless, as will be subsequently shown, when N is even there is an imaginary term in the response. In order to get rid of this term, N must be odd.
MATRIX ~RMULATION

v=!EEHp
2n *

(13)

where E is the matrix defined in eqn (1 I) with positive signs in the exponentiais instead of negative ones, and H is the diagonal matrix formed with the complex frequency response functions calculated at the discrete frequencies of Table Al. The typical term of H is given by

mk)=(k
Let P= M&&~(G),&), and . . . ,P(G), . . . vN,,-,)l (6)

-mrz;5,+ir5&-,

(O<m

<N--l),

(14) where k, m, and c are the stiffness, mass, and damping of the SDOF system, respectively. Substituting now P from eqn (10) into eqn (13), the following equation is obtained v = f EHE*p. (1%

be, respectively, the vectors of the load and the response at the discrete times r, = ndt, and let P = {P(&), P(G,,), P(&), . . . , P(&), ...I (9) (rz = 0, 1,2,. . . , N - l), (8)

Equation (15) expresses the matrix formulation of the dynamic analysis of SDOF systems in the f&quency domain. In the sequel, it is proven that the matrix product EHE*p is rest, provided N is odd. Consider from eqn (2) P&J and P&_,), with written in indiciai notation, m=O,l,...,(N-1)/2 respectively, as ; and

W,) = E:,P,

P(G - 1)I

be the vector of the discrete Fourier transform of the load defined at the discrete frequencies &&interpreted according to Table Al. With the definition of eqns (6) and (9), eqn (2) can be cast in matrix form as P = AtE*P, (!O)

where the (N x N) matrix E* is defined as the matrix whose generic term E& is E& = e -i~&?fflN) or, explicitly (11)

N - 1. All the corresponding where j=1,2,..., terms in the summations of eqns (16) and (17) are complex conjugates (except the first ones which are real) in face of the proof given in Appendix 2. Therefore, P(ui,) and P(G,_,) are complex conjugates. Consider now eqn (13) written as $+4P,

W-0

e e*
e-OWl

. . . eo
.
..e - in(Zn/N)

..eO
. . .

e-.w-

W~/N) 1)(2dYl

. * . e-9Jowl

. , . e-izw-

E*= Symmetric

Technical Note where A is given by A=EH. (19)

855

The n th line of matrix A from eqn (19) can be written as the vector A,= { x H(ci&_nt). . . 1 As (ewZnw, e~N-m)(2n~) and [H(Q,,), H(8,_,)] are pairs of complex conjugates (see Appendices 2 and 1, respectively), the typical pair displayed in eqn (20) is also a pair of complex conjugates. Multiply now A,, from eqn (20), by P, from eqn (9), in order to obtain the typical term of v, u,, from eqn (18). The result, in indicial notation, is ACT XV = &nW&JWrn). In this summation, L%JWL)WJJ, all pairs like (21) (20) $... einm(2Wl H(~~). . . ein(N-mWlN)

optimizing the space of memory in the computer and the number of operations in the process. The matrices E, H and E* have properties that can simplify the numerical implementation of eqn (15), also reducing the computational effort.
REFERI5NCES J. D. Kawamoto, Solution of nonlinear dynamic structural systems by a hybrid frequency-time domain approach. MIT Research Report RS3-5, Department of Civil Engineering, Cambridge, MA (1983). G. R. Darbre and J. P. Wolf, Criterion of stability and implementation issues of hybrid frequency-time-domain procedure for nonlinear dynamic analysis. Transactions of 9th International Conference on Structural Mechanics in Reactor Technology, Lausanne, August (1987).

F. Venancio Filho and A. M. Claret, Non-linear dynamic analysis with frequency-dependent damping.
Damping-89, Florida, February (1989). R. W. Clough and J. Penzien, Dynamics of Structures. McGraw-Hill, New York (1982).
APPENDM 1

The discrete frequencies employed in this formulation must be interpreted according Table Al. Taking into account the frequencies Cs, from Table Al, H(ci&,) and H(e5N_m), eqn (14), are complex conjugate. Table Al. Discrete frequencies (N odd) m
0

En(N-m~H(~N-m)P(~N-m)l, (22)

m or (N-m)
0

%I
0

N - 1 and where c.Gm is interwhere m=l,2,..., preted according to Table Al, are complex conjugates pairs. On the other hand the first term (m = 0) in the summation of eqn (21) is real. Therefore, v, is a sum of a real term (the first one) with pairs of complex conjugates which finally proves that v, is real and, consequently, v is real. A very important point that must be emphasized is that N must be odd, otherwise there would be in the summation of eqn (21), a central term E !l(N/2)
f&&v,2 PY&v72 1

1 2 (N - .1)/2 (N + 1)/2 ... N-2 N-l ;;I

1 2 ;;;;

Afi 2A(i, ... [(N - 1)/2]Adi [-(N - 1)/2]Aci,

. .
2 1

. .
-2Ac5 -Be3

APPENDIX

Apply Euler formula to eimn(**~ in order to have e~2x~M=cos(mn$)+isin(mn$) and to e(~R-m~2rlvJ to have e<N - m)n(Zn/M =cos[(N-m)n;]+isin[(N-m)n%] (A2.1)

which has not its complex conjugate in order to form a complex conjugate pair. In this way, there would exist an imaginary term in the summation which produces v,. This is contrary to the condition implied in FFT algorithm that N should be a power of two and, consequently, even.

FINAL REMARKS

Equation (15) establishes a matrix formula for the analysis of the response of a SDOF system in the frequency domain. The aim of this formulation is to minimize the computational effort required in nonlinear analysis in the frequency domain. On the other hand, in eqn (15) the only condition implied over N is that it should be odd. Thus, N can be changed freely to achieve sufficient precision in the response,

=cos(mn$)-isin(mn2).

(A2.2)

Equations (A2.1) and (A2.2) prove that ebnn(2Rn? and e@v-mM2ni~ are complex conjugates.