You are on page 1of 14

Pergamon PII: F&045-7949(96)00315-X

CompuSrs & Slrucrures Vol. 62, No. 6, pp. 985-998, 1997 Copyright 0 1996 Ekvier Science Ltd Printed in Great Britain. All rights reserved 00457949/97 $17.00 + 0.00

A CF!ITIQUE OF MODE ACCELERATION AND MODAL TIRUNCATION AUGMENTATION METHODS FOR MODAL RESPONSE ANALYSIS
J. M. Dickens, J. M. Nakagawa and M. J. Wittbrodt
Lockheed Martin Missiles and Space, Box 3504, Sunnyvale, CA 94088-3504, U.S.A.
(Received 30 October 1995)

Abstract-The mode acceleration and modal truncation augmentation methods used in modal response analyses are reviewed in detail. The rationale for using the methods is examined and a theoretical relationship between the two methods is developed. The methods are compared with a small numerical example problem by evaluating the dynamic response of a structure for a harmonic loading and a time domain loading. The modal truncation augmentation method is found to be superior to the mode

acceleration method for both implementation and quality of results. Copyright 0 1996 Elsevier Science Ltd.

1. INTRODUCTION

2. MODAL STRUCTURALRESPONSE ANALYSIS

The modal response or mode superposition method for evaluating the response of a structure to either time domain or frequency domain loading is well established and is used extensively in all areas of structural vibration analysis. During a modal response analysis, it is often advantageous or even necessary to reduce a model to a much smaller number of dynamic degrees of freedom prior to solving for the responses. Unfortunately, the reduction process alters the modal representation of the applied loading and can adversely affect the quality of the calculated responses. Two methods that attempt to improve the truncated modal representation of the loading are the mode acceleration (MA) method [l-5] and the modal truncation (MT) augmentation method [6,7]. The mode acceleration method is used primarily by aerospace engineers for coupled loads analyses and has not been used extensively outside of this field of specialization. The modal truncation augmentation method is much newer and is only beginning to be implemented in all fields of structural vibration analysis. The first section of this paper gives an overview of the standard modal response method and details the loss in accuracy as a result of the modal approximation. The second and third sections describe the MA method and the MT method, respectively, and present some insights into both methods. In the fourth section, the methods are compared and the relationship between the two is developed. In the last section, the conclusions from the theoretical comparison are validated with a numerical example of harmonic and time domain loading.

Let the discrete equations of motion structural system be given by:

of a linear

(1)
where M, C and K are the neq x neq mass, damping and stiffness matrices, respectively; y(t), f(r), j;(t) are the physical displacement, velocity and acceleration vectors, respectively; and R(t) is the applied loading vector. R(t) is composed of two parts: Ro is the invariant spatial portion and r(t) is the time varying portion. The loading may be composed of several spatial load vectors each with a corresponding time varying portion. For simplicity, only a single vector is used herein. For a modal response analysis, the physical coordinates of eqn (1) are transformed to modal coordinates, g(t), by a retained set of eigenvectors of the system, 9: {Y(O) =

PlIgw~

(2)

where @ are determined from the general eigenvalue problem:

Klul[@l = w1[@1~~*1.

(3)

The number of retained eigenvectors, given by ip, is typically much less than neq. If eqn (2) is used to transform eqn (1) to modal coordinates:

985

986

J. M. Dickens et al. subtracting the modally represented spatial load vector, R,, from the applied spatial load vector, R,,
W

where [@][KJ[@] = [Q] = diagonal of {wz} n = 1, ip

{R} = {Ra} - {R,} where R, is given by (see Appendix):

(8)

WI Pl[clPl = [Gl
PIT{R~} = {a}. (4c) (4d)

(9)
The response due to R, represents the portion of the overall response that is lost because of the spatial load truncation. If R, is reduced, this loss would decrease and the quality of the overall responses should improve. Obviously, a reduction in the spatial load truncation can be realized by increasing the number of retained modes. However, the number of additional retained modes needed for convergence to an accurate solution can become significant, possibly negating the computational advantages gained by the modal truncation. Fortunately, there are techniques that not only address the problem of spatial load truncation, but also improve the modal efficiency with which the solution converges. The MA and MT algorithms are two such methods. As an aside, an alternate solution procedure is the Ritz Vectors (RV) algorthm [9]. In this algorithm, the spatial load vector is implicitly contained in the reduction vectors used to reduce eqn (1) to a smaller uncoupled set of equations similar to eqn (4). The number of vectors to retain in the RV method is determined by the fidelity to which the frequency content of the loading, given by the term r(t), is to be represented in the solution. The RV method can be considered to be the antithesis of the methods described herein, in that the spatial loading is completely represented at the outset and the frequency content is approximated.

Although not required, a modally diagonalizable damping matrix is commonly assumed and, thus, C, could be rewritten as the diagonal matrix, [2&w.]. This substitution would uncouple eqn (4) into ip separate equations. The standard procedure for determining responses, referred to as the mode displacement (MD) method [4,5,8], consists of expanding the modal responses, solved from eqn (4), into approximate physical responses using the retained modes, @: {Y(0) = {jr(t)} =

PIW~ PlM~)I

(5)

(6)

The solution is referred to as approximate in the sense that the exact solution would include all of the system modes. Although eqns (1) and (4) are written for the time domain, the equations apply equally to the frequency domain. For the frequency domain, the displacement is given by y(f) instead of y(t) with similar transformations for the velocity, acceleration and loading. Thus, while not explicitly identified, the following discussion is also applicable to the frequency domain. In the transformation of the physical equations to the reduced set of modal equations, the choice of the number of modes to retain is typically driven by the need to span the frequency content of interest. That is, the frequency of the highest mode retained should sufficiently exceed the frequency content of the applied time history loading by a predetermined margin. However, this criterion only addresses the time dependent portion of the applied loading, r(t) and ignores the spatial portion, R. Undesirable inaccuracies in the response calculations can occur by not considering the effects of modal truncation on the spatial portion of the applied load. To quantify the amount of the spatial load truncation, let the portion of the load vector not represented by the retained modes be the force truncation vector, R,. Note that the response due to R, will be quasi-static since, by definition, the significant dynamic response has been enveloped by the retained modes. R, can be determined by

3. MODE ACCELERATION METHOD

The mode acceleration method is well documented [l-5]. The logic of the MA method is that since, by definition, the number of modes retained accurately spans the frequency range of interest, any loading represented by the non-retained modes will produce a quasi-static response. Consequently, the response due to the non-retained modes will have no dynamic amplification. That is to say, the modes not retained will cause no appreciable velocity or acceleration response and, thus, the physical accelerations and velocities can be determined directly from their modal counterparts as in the MD method. Then, since all of the quantities are now known, eqn (1) can be used to calculate the physical displacements. To develop the MA algorithm, eqn (1) is rewritten as

Augmentation methods for modal response analysis or as {R} = {E}r(r) - {Ri) - {RI) (lob)

987

where R, Ri and & are the elastic, inertia and damping forces, .respectively. Using eqn (2), the inertia and damping forces can be rewritten in modal coordinates as: Pi) = [W[@l{ii(t)}

number of spatial loads of eqn (1) plus twice the number of modes retained. For eqn (16), the time dependent DOFs is equal to the number of spatial loads plus the number of retained modes only. Typically, the number of spatial loads is much smaller than the number of retained modes. Since modally diagonalizable damping is normally assumed, the following simplification can be made to the damping term of eqn (13):

(11) (12)

Ki-[CT@]= Kl-'P41[@1[2Lwnl n = 1, ip (17)


where [2l.o,,] is a diagonal matrix. Thus, eqn (13) can be reduced to the following equation, which has the same number of time dependent DOFs as eqn (16). {Y$)} = [fl-{Ro)r(t) - WPflPlW)~

{Rd) = [cl[@l{k(t)).

Finally, using eqns (10X12), the displacements using the MA algorithm, y.(t), are given by: {y&)1 = K-P01r(t) - Kl-Vfl[@l{~(~)~

-K--KlPlM~)I = [% g

(13)

where

The MA physical velocities and accelerations, jr.(r) and j;,(t), respectively, are determined using eqns (6) and (7). {jil&)) = Pwm {Y&)1 = PIW)). (14)

(15)

If the damping force is assumed to be negligible, the last term of ecln (13) can be dropped resulting in the following for the MA displacements.

(16)

The MA method of the form of eqn (16) to determine the physical displacements was originally proposed by Williams [2] and was reported by Bisplinghoff et al. [I] and Bisplinghoff and Ashley [3]. It is not clear from these references if eqn (16) was recommended for use in all cases or for only the undamped case. I-lurty and Rubenstein [4] derive the method in the form of eqn (13) only and the simplification to the undamped case is left to the reader, Craig [5] presents eqn (16) for the undamped case and eqn (13) for the damped case. It is not clear to the authors why the standard implementation wit;hin the industry is to use eqn (16) for damped cases. Possibly this is due to the fact that the time dependent portion of eqn (13) has nearly twice the number of DOFs than eqn (16). The number of time dependent DOFs of eqn (13) is the

In eqn (18) g(t) is the modal force due to the modal acceleration (inertia force) and modal velocity (damping force). Noting that the modal acceleration, g(t), is numerically equivalent to the modal inertia force since the modes on which it is based, @ of eqn (4) are mass normalized to unity. Thus, eqn (18) can be developed by expanding the modal dynamic forces, the inertia forces (g(t)) and the modal damping forces ([2&0.]{g(t))), to the physical domain by the term [w[@] as shown in the Appendix. Another possible justification for using eqn (16) instead of eqn (13) or (18) is the belief that the damping forces are negligible. In many instances, this has not been the experience of the authors, particularly in large models on the order of thousands of physical DOFs where the error of 5-10S is found for many quantities and a few having errors even larger. It is not difficult to construct a small problem in which the error is on the order of 10%. A more likely justification is that many analyses within the aerospace community consist of the assemblage of several dynamically reduced substructures. The recovery of the displacements on the substructure level or at the assemblage level using eqn (13) is considerably more complicated than using eqn (16). There arise many issues such as: (a) what damping forces should be used and at what level, assemblage or individual substructure, are they to be applied; (b) how are the damping forces to be calculated for the boundary of the substructure; and (c) how many DOFs are required for displacement type recovery.

988

J. M. Dickens et al. The MT vectors are appended to the retained modal set, [@I, to construct the pseudo modal set, [&:] [@]-[&I = [@ P]. (24)

4. MODAL TRUNCATION AUGMENTATION METHOD

The modal truncation augmentation (MT) method [6,7] attempts to correct for the inadequate representation of the spatial loads in the modal domain by creating additional. pseudo eigen or MT vectors to include in the modal set for the response analysis. The terminology of pseudo eigen is used because the MT vectors are orthogonal on the mass and stiffness matrices but do not satisfy the eigenvalue problem at each equation. That is, the MT vectors only satisfy eqns (4a, b) but do not satisfy eqn (3). The MT vectors are created using a mathematically consistent Rayleigh-Ritz approximation where the assumed Ritz basis vectors are derived using the spatial force truncation vector, R, of eqn (8). The MT vectors are orthogonal to the retained eigenvectors since the force truncation vector does not contain any components of the retained eigenvectors. For the response solution, the MT vectors and associated Rayleigh-Ritz frequencies are appended to the retained eigenvectors and the modal response analysis proceeds as if the augmented vector set were all eigenvectors. An MT vector is determined by first solving for the displacement vector, {X), due to the force truncation vector of eqn (8) . KIW Form 14 = IXWHX] WI = {X]TWl(X) and solve the reduced eigenvalue problem I@ = MQti;. The MT vectors, P, can now be formed {P> = POQ. (23) (22) (20) (21) = WI. (19)

For only one spatial load vector, the right-handside of eqn (19) will have dimensions of physical DOF by 1 and eqns (20)-(22) will have dimensions of one by one. For say five load vectors, eqn (19) will have dimensions of physical DOF by five, eqns (20H22) will have dimensions of five by five and five additional vectors will be added to the modal set in eqn (24). After completing the modal response analysis, the physical acceleration, velocity and displacement responses are found using eqns (5X7), exactly as in the standard MD method, with the pseudo modal set of vectors.

5. THEORJZTICAL COMPARISON OF THE METHODS

To compare the MA and MT methods, the exact solution of eqn (1) is written in two parts. {Y(t)> = {Y&)> + {YtW. (28)

The first part, y,(t), is the portion of the displacement solution obtained from the retained modes and the second part, y,(r), is the portion of the displacement solution lost due to the modal truncation. Substituting eqns (28) and (8) into eqn (1) Wll(ji&)I + ]WsW] + WHY&)} + WHY&)] (29)

+Wl{j;,(r)] = {R>+ R,}r(t).

+ tcl{W)f

t+

w9 or RQ WY4 h

WY3

m = 1.0, k = 10,ooo. Fig. 1. Example numerical problem model.

Augmentation methods for modal response analysis Table 1. Complete example eigen solution
Mode Frequency no. (Hz)

989

DOF no. 2 3

1
2 3 4

10.155 20.222 28.258 34.963

0.39948 0.63631 0.61408 0.034183 0.68548 0.26428 -0.58359 -0.48927 0.60461 -0.69678 0.19839 0.46815 0.07056 -0.19939 0.49292 - 1.1936

The modal responses solved from eqn (4) yield a solution in the physical domain that satisfies the equation: WlW)1 +

PWsWI + KIIY&)} = {B~)r(t).


(30)

Subtracting WlW)>

eqn (30) from eqn (29) + IWt

(4 + WHY&N = {B,)r(G.
(31)

Equation (31) is the portion of the solution not represented by the modes retained in the analysis. Both the MA and MT methods attempt to find a solution for eqn (31) without calculating the nonretained modes. The difference between the two methods is in the approximation for y,(r). For the MA method, let the solution, y.(t), be partitioned similar to eqn (28) {Y.(0) = {YS(N + {Y&N. (32)

Substituting eqns (32) and (8) into eqn (13) and using eqn (30), the MA solution for the displacements contained in the non-retained modes is given by Kli:~Jt)) = {RN). displacements (33) are

For the MT method, the partitioned similarly given by

{y..(r)) = {YSW>+ {Yl.(0).

(34)

The solution not represented by the modes, y,,(r), is created using eqns (19)-(24). {y,(t)) = Using this transformation

v%m~.
to reduce eqn (31):

(35)

[~lTwmmowj

+ [~l[afwm~

+ [~lTLKIWz,&)~ = PITPt b(t). (36)

Equation (36) can be solved for g+,(t) in the same manner that eqn (4) was solved for g(t). g,(t) can then be back transformed using eqn (35) to yield the MT solution for the displacements contained in the non-retained modes. Both the MA method and MT method contain the same modally retained portion of the response, y,(t), as given by eqns (28), (32), and (34). For the non-modally represented portion of the solution, y,(t), the solutions are different. The MA method approximates y,(t) by a static solution, y,(t), which lacks any dynamics. The MT method approximates y,(t) by a dynamic solution, y,,(t), which averages all non-retained modes by a truncation augmentation vector, P, in a consistent dynamic equation, eqn (36). It should be noted that for both methods, the displacement static solution is recovered. If the acceleration and velocity of the MT method due to the non-retained modes are set equal to zero, the solution for the displacements will be equal to the MA method. This results since the vector P of the MT method is a linear multiple of the term [ICI-{R,} of eqn (33) (or a linear combination in the case of multiple loading vectors). The MA method is an approximation of the MT method that is obtained by forcing the response of the non-retained modes to be static. The computer implementation of both methods provides another point of comparison. The MA method requires extensive modifications to the standard mode displacement solution algorithms. In addition, the modifications are different depending upon whether a time domain, frequency sine sweep, random response, or shock spectrum type of modal response analysis is being performed. Conversely, the MT method requires no modifications to the standard modal response computer routines for either time domain or frequency domain, as well as, shock spectrum type analysis. The physical quantities are obtained in the same manner as the standard MD method. The implementation of the method only requires calculation of the truncation vectors and frequencies that are appended to the modal set. In summary, it is noted that: (a) the MA and MT method both approximate the non-modally represented solution; (b) the MA method is an approximation to the MT vector method; (c) due to the added dynamics, it would be expected that the MT method would give better results overall than the MA method; and (d) the MT method is simpler to implement with no modifications to the standard MD method computer routines. The numerical example in the next section confirms the first three points.

Table 2. Modal truncation vectors Mode no. 1 MT mode Frequency 0-W 10.155 21.865 DOF no.

I
0.39948 -0.53176

2 0.6363 1 -0.41943

3 0.61408 0.71885

4 0.34183 0.22164

990

J. M. Dickens et al.

3. R(f) = W(f)

70.

Frequency (Hz)

0.

L
2.0 r -1.0

.030 .033

Time (set)

R(t) = QW
1. o.ooo 5 t 5 0.030

r(f) = 1.

r(t) = i 0. 0.033 5 t 5 1.500 Fig. 2. Schematic of frequency domain and time domain loading.

3.5

f 5 70.

6. NUMERICAL COMPARISON OF THE METHODS

To present a numerical comparison between the MA and MT methods, a harmonic response (sine sweep) analysis and a time domain response analysis were performed on a simple four DOF model. For the analyses, both methods retained a single mode and determined the physical accelerations, displacements, and spring forces. To provide a baseline for the comparison, the analyses were also performed using the mode displacement method while retaining all modes of the system to give the exact solution. The four DOF model created is shown in Fig. 1. For the model, the assembled mass and stiffness matrices are:

-1.0 2.0 -1.0 -1.0 2.0 -1.0 -1.0 2.0

K = 10,000.

In order to have four distinct modes in the example, the mass at DOF 4 was set to half the value of the other three DOFs. The modal damping is 2% (Q = 25) for all modes and the MT vector.

I 10 Fw-nc~ Ml

1 70

Fig. 3. Displacement amplitude response for DOF 3 using all modes, MA method and MT method.

Augmentation methods for modal response analysis

991

......

MA Method

10

70

Frequency

[Hz]

Fig. 4. Spring force amplitude response for spring 4 using all modes, MA method and MT method.

The spatial loading, &, consists of a single force at DOF 3.

physical force for the eigenvector can be shown to be:

Spatial loading of first mode =

The complete eigen solution for the problem is given in Table 1. Only one eigenvector was retained for both the MA and MT methods. Using the method outlined in the Appendix, the equivalent

The MT method vectors (the first eigenvector and the MT vector) are given in Table 2. The MA method used eon . (16) . I to recover the displacements.

2x10 7
1x10': ;; 8 t Q "5 1x10"' !I 4 1x100:

10

70

_FW-cy WI Fig. 5. Acceleration amplitude response for DOF I using all modes, MA method and MT method.

992

J. M. Dickens et al. (x10, AllModes MA Method % error Ml Method % error

DOFl 4.52 4.50 1.1% 4.54 0.3%

DoF2 8.89 8.92 0.3% 8.89 0.0%

DoF8 12.90 12.90 -0.4% 12.90 0.0%

DOF4 6.53 6.49 -0.5% 6.51 -0.3%

DOFl HALL

DOF2

DOF3

DOF4

q MA WMT

Fig. 6. Summary of displacement at 3 Hz for the harmonic response example.

7. HARMONIC

RESPONSE ANALYSIS

The harmonic response analysis consists of computing the steady-state response of the mode1 as a function of the frequency of applied harmonic force. The applied frequency domain loading is given in Fig. 2. Typical responses of the problem from 3.0 to 70.0 Hz for displacement, spring force and

acceleration are shown in Figs 3-5. In addition, Figs 68 present the amplitude response at 3.0 Hz for all DOFs and the percentage error compared to retaining all the modes. For the displacements, Figs 3 and 6 show that both the MA and MT methods produced nearly the same results for the frequency range from 3.0 Hz to the first mode of 10.2 Hz. From Fig. 6, it is seen that the

Spring 1 All Modes MA Method % error MT Method % error 0.45 0.46 1.1% 0.45 0.3%

Spring 2 0.44 0.43 -0.5% 0.44 -0.3%

SPmI3 0.41 0.40 -2.0% 0.41 0.0%

SPml4 0.64 0.64 -0.3% 0.64 0.3%

Spring 5 0.65 0.65 -0.5% 0.65 -0.3%

Spring 1
HALLCIMAHMT Fig. 7.

Spring 2

Spring 3

SPWI 4

sprlno 5

Summary of spring force at 3 Hz for the harmonic response example.

Augmentation methods for modal response analysis


I

993
DOF4 2.32 2.01 -13.5% 2.31 -0.3%

(x 10-q All Modes MA Method %error MT Method I % errrx 8 2 xc 2 z 6.0

DOFl 1.61 2.34 45.9% 1.61 0.3%

DOF2 3.16 3.73 16.3% 3.15 0.0%

DOF8 4.59 3.60 -21.6% 4.59 0.0%

s E e 0 3 8
a

4.0 x p (D 2.0 s
0.0 DOFl DOF2 DOF3 DOF4

MALL q MA HMT

Fig. 8. Summary of acceleration at 3 Hz for the harmonic response example, maximum absolute percentage error for the MA method at 3.0 Hz is 1.1% while the MT method maximum absolute error is 0.3%. Over all the DOFs, the MT method is consistently closer to the exact solution, although both methods had errors of less than 2.0%. The small errors are to be expected since 3.0 Hz is basically static compared to the first mode of 10.2 Hz and both methods recover the full static displacement. For the spring force, Figs 4 and 7 show that both methods again produced nearly the same results for the frequency range from 3.0 to 10.2 Hz. From Fig. 7, it is seen that the maximum absolute percentage error for the MA method at 3.0 Hz is 2.0% while the MT method maximum absolute error at 3.0 Hz is 0.3%. These results appear more favorable to the MT method than were determined for the displacements. Although the sample size is small, these results make sense since the spring forces are based on relative displacements and thus, overall, one would expect that errors in the MA displacements would only be magnified in the spring forces.

.......

MA Method

8 I I

i D
-1.Ox10~--+
I I I I I I I I I

0.1

0.2

0.3

0.4

0.5
Tlmo [aec]

0.6

0.7

0.6

0.9

Fig. 9. Displacement response for DOF 1 using all modes, MA methods and MT method.

994
0.5

J. M. Dickens et

al.

.......

All Modes

MAMethod

i 3 i

03 .

gg 0.0

E u g -0.3 E $

-0.5
0 0.1

0.2

0.3

0.4

0.5 0.6 Time [me]

0.7

0.8

0.9

Fig. 10. Spring force response for spring 3 using all modes, MA method and MT method.

A comparison of the acceleration response, Figs 5 and 8, shows that the MT method is a good approximation to the exact solution whereas the MA method had significant errors. At 3.0 Hz, the MA method has errors as high as 45.9% whereas the MT method had only a maximum error of 0.3%. The significant acceleration error in the low frequency range can be explained by noting from eqn (36) that the truncated force captured by the MT vector includes dynamic effects from the non-retained modes, including acceleration effects. In the MA method, only displacement is created by the truncated forces as shown in eqn (33).

Although neither method was developed for accurately calculating the response for the nonretained modes, the difference in the two methods is seen best by looking at the response above the modes retained, (in this example 10.2 Hz). As shown in Figs 3-5, the dynamic response for the MA method above the highest natural frequency is static, as illustrated by the asymptotic response of the displacement, spring force and acceleration. In contrast, the MT method attempts to average, in a consistent mathematical sense, the response of all the modes not retained in the analysis into a single MT mode. In other words, the MT method averages with

1.0

....... 2 8 0.5

All Modes MA Method

n! 5 H 0.0

I u

-0.5

-1 .o

0.2

0.3

0.4

nme [aoc]

0.5

0.6

0.7

0.8

0.9

Fig. 11. Acceleration response for DOF 3 using all modes, MA method and MT method.

Augmentation methods for modal response analysis

995 DDF4 8.72 4.54 -32.4% 4.99 -25.7%

104

DDFI 3.84 5.97 55.4% 5.42 41.0%

DDF2 7.81 9.50 21.8% 8.10 3.6%

DOF 3 9.48 8.18 -13.9% 9.83 3.7%

All Modes MA Method % error MT Method % error 15.0 MS a r E s E s Z g= P e 4 g P PO 10.0 5.0 0.0

BALL q MA Ht.lT

(x 16d, All Modes MA Method %Wor MT Method

DOFl -8.08 -5.65 -30.1% -7.33 -9.3%

DOF2 -9.52 -8.99 -5.6% -10.30 7.6%

DOF 3 -7.16 -8.68 21.% -8.21 14.7%

DOF4 -4.42 -4.83 9.3% -4.34 -1.8%

c
a

% error -12.0

ps p

-4.0

% -2.0 0.0
DOFl DOF2 DOF3 DOF4 OMA

HALL

n MT

Fig. 12. (a) Maximum positive displacement summary for the time domain example. (b) Maximum
negative displacement summary for the time domain response example.

a dynamical equation, eqn (36), the response of all the higher non-retained modes while the MA method forces the higher modes to respond only statically, eqn (33). It can also be seen that the MT method did reasonably predict the magnitude of the maximum response in the higher frequency range.

8. TIME

DOMAIN RESPONSE ANALYSIS

The time domain response analysis consists of computing the response of the model to a specified applied force time history. For this analysis, the applied force is a step input as shown in Fig. 2. This type of input tends to excite all the modes of a

996

J. M. Die :kens

et al.

structure since the step function contains all frequencies. Although both the MA and MT methods are not designed for analysis with the loading frequency content above the highest retained mode, a time domain loading satisfying this criterion would not add anything new to the comparison. In addition, typical time history loading is composed of piece-wise linear loads of which the limit for a single step would be a step response. For these reasons the step input was chosen to give the severest comparison of the (a
Spring 1 All Modes MA Method 0.38 spring 2 0.44

methods. In the time domain analysis, a step size of 0.003 s is used. Typical time history response curves for displacement, spring force and acceleration are given in Figs 9-l 1 for the time interval to 1.Os. These curves are included to point out some salient points in the comparison of the methods, however, since the extreme amplitudes are of primary interest, Figs 1214 present the maximum positive and negative response for all DOFs and the percentage error

Spring 8 0.28

SpWb 4 0.48

Wng 0.44

0.80 66.4%
0.64 41 .O%

0.32 -29.1%
0.34 -24.2%

0.03 68.8%
0.41 46.6%

0.39 -19.6%
0.40 -17.5%

0.48 9.3%
0.43 -1.6%

% error
MT Method % error

Spring 1 k#A;LOMA~MT

Spring2

Spring3

Spring4

Wng 6

Spring 1 All Modes -0.81

sprlns2 -0.32

Spdng 3 -0.31

SpW -0.41

Ww6 -0.67

a 3s IE !j % ,g Q

-1.0 -0.6 -0.6 -0.4 -0.2

Spring 1

Spring2

Spring3

qALLOMAUMT

Spring4

spring 6

Fig. 13. (a) Maximum positive spring force summary for the time domain response example. (b) Maximum negative spring force summary for the time domain response example.

Augmentation methods for modal response analysis )


All Modes MAMethod % error MTMethod % orror 1.0 @ I= 2 3 0.6 f a 5 0.6 DDFl 0.66 0.23 DOF2 0.53 0.37 DOF3 0.46 0.35 -23.7% 0.56 21.1% DOF4 0.77 0.20 -74.3% 0.23 -69.4%

991

-65.4%
0.56 -15.6%

-30.4%
0.63 16.6%

e 2 f 0.4 sp 84 0.2 0.0


HALL OMA WMT

._
I.

1)
AllModes MAMethod % orror MTMethod % error
I.

DOFl -0.39 -0.24 -37.5% -0.40 3.1%

DOF2 -0.46 -0.39 -16.6% -0.44 -6.0%

DOF3 -0.96 -0.37 -61.2% -0.67 -9.5% s

DOF4 -0.60 -0.21 -74.1% -0.36 -54.9%

aALL

q MA WMT

DOFl

DOF2

DOF3

DOF4

Fig. 14. (a) Maximum positive acceleration summary for the time domain response example. (b) Maximum negative acceleration summary for the time domain response example. compared to retainmg all the modes. These results are for the times greater than 0.030 s, the interval in which the response has peaked after the initial motion from the static offset initial condition. The displacement response of DOF 1 is plotted in Fig. 9 and shows that the displacements are mainly driven by the first mode with all three curves oscillating at basically the same frequency. However, the higher frequencies can have some significant effects as shown by the all modes curve extending further in the negative direction and being cut off in
CAS 62/6-C

the positive direction. Due to the added dynamics of the MT vector, the MT method was able to capture the extension in the negative direction while the MA method could not. Both methods were unable to match the dynamics in the positive direction, exceeding the exact solution. Figures l2(a, b) show the MT method is generally better at matching the maximum responses with only the maximum negative response of DOF 2 slightly favoring the MA method. For the spring forces, Fig. IO was selected to show what is very close to the limiting case when

998

J. M. Dickens et al. 4. W. C. Hurty and M. F. Rubinstein, Dynamics of Structures. Prentice-Hall, Englewood Cliffs, NJ (1964). 5. R. R. Craig, Jr, Structural Dynamics. Wiley, New York (1981). 6. J. M. Dickens and E. L. Wilson, Numerical methods for dynamic substructure analysis, EERC-80/20. University of California, Berkeley, CA (June 1980). 7. J. M. Dickens and K. $. Pool,Modal truncation vectors and periodic time domain analysis applied to a cyclic symmetry structure. Comput. Struct. 45(4), 685-696
(1992).

comparing the two methods. What can be gathered from this figure is that there are virtually no spring force dynamics contained in the single retained mode for spring 3. Thus, the MA method could only capture the quasi-static correction of the nonretained modes as shown by the near step response with very little overlaid dynamics. The MT method, however, includes the dynamics of the MT vector and is shown as the response amplification that overshoots the exact solution. Overall, Figs 13(a, b) again show that the MT method is better with only the maximum negative responses of springs 2 and 4 slightly favoring the MA method. For the acceleration response of DOF 3, Fig. 11 shows that the MT method exhibits the higher frequency response due to the MT vector that is not contained in the MA method. Figures 14(a, b) show that all maximum responses clearly favor the MT method. From the time domain response analysis, it would appear the MT method outperforms the MA method. Even though the applied force history chosen for the numerical evaluation was a limiting case and the example problem is fairly simple, the experience of the authors on more expansive problems corroborates this conclusion.

8. L. Meirovitch, Elements of Vibration Analysis. McGraw-Hill, New York (1975). 9. E. L. Wilson, M. W. Yuan and J. M. Dickens, Dynamic analysis by superposition of Ritz vectors. Earthq. Engng Struct. Dyn. 10(6), 813-823 (Nov. 1982).
APPENDIX

Modal-physical

representation of displacements and loads

Given the eigenvalue problem [E(1{@.}= and the statics problem [E(l{YI= {RoI with physical displacements in terms of modal displacements as

Wl{@}d

9. CONCLUSION

Then for a solution of the static problem on a mode by mode basis is {@n}T[fl{@.}g. = {@.}T{R~} w.g. = {@.jT{Ro} = u.

The mode acceleration method and the modal truncation augmentation method are compared analytically and numerically. The mode acceleration method was shown to be an approximation to the modal truncation augmentation method. The numerical implementation of the modal truncation augmentation method is simpler than for the mode acceleration method since existing mode displacement response analysis computer routines can be used without modification. For the application of both methods in the designed environment (frequency content of the loading below the highest retained mode), both methods yield nearly the same results for displacements and spring forces with the modal truncation augmentation method being slightly better; however, for accelerations, the modal truncation augmentation method is superior. The benefits are clear, the modal truncation augmentation method is recommended.

g. =
The physical force, by:
{R.},

represented by each mode is found

{R,} = [Kj{k}g. = [k]{@.}u&. = [~I{@.}&~ {R,} = [M]{@,}u,.


For completeness the modal from physical and physical from modal forms for displacement and forces are listed below: (1) Physical displacement, {y.}, from the nth modal displacement, g: {Yn)= I@, I& (2) The nth modal displacement, g., from a physical displacement, {y}:

REFERENCES

R. L. Bisplinghoff, H. Ashley and R. L. Halfman, Aeroelns&ity.Addison-Wesley, Reading, MA (1955). D. E. Williams, Dynamic loads in aeroolanes under given impulsive loads with particular reference to landing and gust loads on a large flying boat. Great Britain Royal Aircraft Establishment Reports, SME 3309 and 3316 (1945). R. L. Bisplinghoff and H. Ashley, Principles of Aeroelasticity. Dover Publications, New York (1962).

& = {@.ITIMl{YI (3) The nth modal force, a., from a physical force {Ro}:

a. = {@.}~{RcI}
(4) The physical force, a.:
{R,},

from the nth modal force,

{R.} = [M]{@~}a~.

You might also like