Fdocuments - in - The Complete Srss Modal Combination Rule PDF

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS

Earthquake Engng Struct. Dyn. 2011; 40:1181–1196


Published online 7 December 2010 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/eqe.1082

The complete SRSS modal combination rule

Ernesto Heredia-Zavoni∗, †
Instituto Mexicano del Petróleo, Eje Central Lázaro Cárdenas 152, México D.F., México 07730, Mexico

SUMMARY
The complete Square-Root-of-Sum-of-Squares (c-SRSS) modal combination rule is presented. It expresses
the structural response in terms of uncoupled SDOF modal responses, yet accounting fully for modal
response variances and cross-covariances. Thus, it is an improvement over the classical SRSS rule which
neglects contributions from modal cross-covariances. In the c-SRSS rule the spectral moments of the
structural response are expressed rigorously in terms of the spectral moments of uncoupled modal responses
and of some coefficients that can be computed straightforwardly as a function of modal frequencies and
damping, without involving the computation of cross-correlation coefficients between modal responses.
An example shows an application of the c-SRSS rule for structural systems with well separated and
closely spaced modal frequencies, subjected to wide-band and narrow-band excitations. Comparisons
with response calculations using the SRSS and the Complete Quadratic Combination rules are given and
discussed in detail. Based on the c-SRSS rule a response spectrum formulation is introduced to estimate
the maximum structural response. An example considering a narrow-band excitation from the great Mexico
earthquake of September 19, 1985, is given and the accuracy of the response spectrum formulation is
examined. Copyright 䉷 2010 John Wiley & Sons, Ltd.

Received 26 November 2007; Revised 8 October 2010; Accepted 11 October 2010

KEY WORDS: modal combination rules; modal analysis; seismic response analysis; SRSS rule; CQC
rule; response spectrum

INTRODUCTION

An established procedure for the seismic response analysis of linear MDOF structures is to uncouple
the equations of motion and transform them into a series of SDOF modal oscillator equations
whose maximum responses can be readily computed. The maximum structural response can then be
obtained from the SDOF modal maxima using modal combination rules. A widely used combination
rule is the Square Root of the Sum of Squares (SRSS), introduced by Rosenblueth [1], where
the maximum structural response is estimated as the square root of the sum of squared modal
maximum responses. The rule takes basis on expressing the structural response variance as the
sum of modal response variances, thus neglecting the cross-covariance between modal responses.
It is known that this may be appropriate for structures with well-separated modal frequencies
subjected to wide-band ground motions. However, for closely spaced modal frequencies the cross-
covariance between modal responses becomes more significant and cannot be neglected in the
estimation of the response. Rosenblueth and Elorduy [2] suggested an improved rule that attempted
to account for interaction between modal contributions. Der Kiureghian introduced the Complete
Quadratic Combination (CQC) rule [3–5], where simple expressions for the correlation coefficients

∗ Correspondence to: Ernesto Heredia-Zavoni, Instituto Mexicano del Petróleo, Eje Central Lázaro Cárdenas Norte
152, México D.F., México C.P. 07730, Mexico.
† E-mail: eheredia@imp.mx

Copyright 䉷 2010 John Wiley & Sons, Ltd.


1182 E. HEREDIA-ZAVONI

between modal responses are derived based on a white-noise model of the ground acceleration.
This approximation works well when the ground motion is wide banded, with smoothly varying
power spectral density over all significant modal frequencies of the structure. However, for narrow
bandwidth seismic inputs the CQC approximation may not estimate the response accurately. Based
on an approximate solution for white-noise ground motion input, Vanmarcke [6, 7] also proposed
a modal combination rule that expresses the cross-covariance between modal responses as a sum
of modal response variances. Extensions of combination rules for multi-support structures have
also been developed [8–10].
In this paper, the complete SRSS rule (c-SRSS) is formulated. First, it is shown that the structural
response can be expressed rigorously in terms of uncoupled SDOF modal responses accounting
fully for both modal response variances and cross-covariances. An expression is derived for the
spectral moments of the response as a sum of spectral moments of SDOF modal oscillator responses.
It is then shown that the response variance can be expressed as the sum of uncoupled modal
displacement variances multiplied by coefficients which can be computed in a straightforward
manner as a function of modal frequencies, damping ratios and modal characteristic frequencies.
An example is given next showing that the c-SRSS rule provides the correct response for systems
with separated or closely spaced modal frequencies subject to wide-band and narrow-band seismic
inputs. The results are compared against the SRSS and CQC rules. Finally, based on the c-SRSS
rule a response spectrum formulation is introduced to estimate the maximum structural response.
This formulation is applied to an example of narrow-band ground acceleration. The accuracy
of the response estimates is examined and also compared with that of other response spectrum
formulations.

STRUCTURAL RESPONSE

Consider a linear MDOF structural system subjected to earthquake ground acceleration a(t). The
equation of motion of the system can be written in matrix form as follows:
MẌ+CẊ+KX = −MIa(t) (1)
where M, C and K are the mass, damping and stiffness matrices, respectively, X is the vector
of nodal response displacements relative to the ground and I is a transfer vector relating the
displacements of the structure with a unit ground displacement. Let U = [u1 u2 . . . un ] be the
modal matrix, where ui , i = 1, 2, . . .,
n,n are the mode-shape vectors and n is the number of degrees
of freedom. Introducing X = UZ = i=1 ui z i (t), where Z = {z 1 , z 2 , . . ., z n } is the vector of general
coordinates z i (t), and provided the system is classically damped, the equations of motion can be
transformed into n uncoupled equations of SDOF systems with modal properties
z̈ i +2i i ż i +i2 z i = −i a(t) (2)
where i and i are the modal frequencies and damping ratios, respectively, and i = uiT MI;
i = 1, 2, . . ., n, are the modal participation factors. A generic response of interest, Y (t), e.g. internal
force in a member, displacements at a node, or stress at a point, in general can be expressed as
a linear combination of nodal displacements, Y (t) = QT X, where the response transfer vector Q
depends on the geometry and stiffness of the structure. The structural response is then

n
Y (t) = QT ui z i (t) (3)
i=1

For a zero-mean stationary random ground acceleration, it can be shown from (2) and (3) that the
one-sided power spectral density of the stationary response, G(), is given by

n 
n 
n
G() = ci2 |Hi ()|2 G o ()+ ci c j Re[Hi ()H j (−)]G o () (4)
i=1 i=1 j=1
j=i

Copyright 䉷 2010 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2011; 40:1181–1196
DOI: 10.1002/eqe
THE COMPLETE SRSS MODAL COMBINATION RULE 1183

where ci = QT ui i , G o () is the ground acceleration one-sided power spectral density, Re[ ]
denotes real part and Hi () is a SDOF modal transfer function,
1
Hi () = − (5)
i2 −2 +2i i 
From (4), the response variance is given by
 ∞  ∞

n n 
n
Y = ci
2 2
|Hi ()| G o () d+
2
ci c j Re[Hi ()H j (−)]G o () d (6)
i=1 0 i=1 j=1 0
j=i

The single and double sums in (6) are the contributions from the modal response variances and
from the cross-covariance between modal responses, respectively. If the latter are neglected, then
 ∞

n
Y ≈ ci
2 2
|Hi ()|2 G o () d (7)
i=1 0

which is the basis for Rosenblueth’s SRSS modal combination rule.

THE COMPLETE SRSS MODAL COMBINATION RULE

It follows from (5) that:


4 +(4i  j i  j −i2 −2j )2 +i2 2j
Re[Hi ()H j (−)] = (8)
[(i2 −2 )2 +(2i i )2 ][(2j −2 )2 +(2 j  j )2 ]

The right-hand side of (8) is a proper rational function of frequency and its denominator is equal
to the inverse of |Hi ()|2 |H j ()|2 ; thus, it can be expressed in terms of the norms |Hi ()|2 and
|H j ()|2 by means of a partial fraction decomposition as follows:

2 2
Re[Hi ()H j (−)] = Aij |Hi ()|2 + Bi j |Hi ()| 2
+ D ij |H j ()| 2
+ E i j |H j ()|2 (9)
i2 2j

Substituting |Hi ()|2 and |H j ()|2 into (9), and equating the numerator to that in (8), one obtains
a system of equations for factors Aij , Bij , Dij and E ij :

Aij 4j + Dij i4 = i2 2j (10a)

4j i4
−2Aij 2j +4Aij 2j 2j + Bij −2D ij i
2
+4D ij  2 2

i i + E ij = 4i  j i  j −i2 −2j (10b)
i2 2j

2j 2j i2 2 i


2
Aij −2Bij +4Bij 2j + D ij −2E ij +4E ij i 2 =1 (10c)
i2 i2 2j j

1 1
Bij + E ij 2 = 0 (10d)
i
2 j

Let q = i / j ; solving the system of equations (10) it is found that

[4 j q(i − j q)+q 2 −1](1−q 4 )−2q 2 (1−q 2 )[q 2 (1−22j )−(1−2i2 )]


E ij = (11)
4q 2 [(1−2i2 )q 2 −(1−22j )][(1−22j )q 2 −(1−2i2 )]−(1−q 4 )2

Bij = −q 2 E ij (12)

Copyright 䉷 2010 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2011; 40:1181–1196
DOI: 10.1002/eqe
1184 E. HEREDIA-ZAVONI

8 8
ξ=0.02 6 ξ=0.02
6
ξ=0.05 ξ=0.05
4 4
ξ=0.10
2 ξ=0.10 2
Aij

Bij
0
-2 -2
-4 -4
-6 -6
-8 -8
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
(a) q (b) q
8 8
6 ξ=0.02 6 ξ=0.02

4 4 ξ=0.05
ξ=0.05
ξ=0.10
2 ξ=0.10 2

Eij
Dij

0 0
-2 -2
-4 -4
-6 -6
-8 -8
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
(c) q (d) q

Figure 1. Partial fraction factors: (a) Ai j ; (b) Bi j ; (c) Di j ; and (d) E i j .

4i  j q −q 2 −1+2q 2 (1−22j )−(q 4 −1)E ij


Dij = (13)
2q 2 [q 2 (1−22j )−(1−2i2 )]

Aij = q 2 −q 4 Dij (14)

where Dji = Aij , E ji = Bij ; Aij = 0.5 and Bij = 0 when q = 1. Notice that factors Aij , Bij , Dij and
E ij depend only on the ratio of modal frequencies, q = i / j , and on modal damping ratios, i ,
 j . The variation of these factors with ratio q is shown in Figure 1 for different i and  j . The
partial fraction decomposition of Re[Hi ()H j (−)] is compared in Figure 2 to the exact values:
it is seen that the partial fraction decomposition yields the correct values of Re[Hi ()H j (−)].
Using (9) in (4), the cross-terms in the response power spectral density can be written as

n 
n 
n 
n
ci c j Re[Hi ()H j (−)] = ci c j [Aij + Dji ]|Hi ()|2
i=1 j=1 i=1 j=1
j=i j=i
 

n 
n Bij E ji
+ ci c j + 2 |Hi ()|2 (15)
i=1 j=1 i2 i2
j=i

and thus

n 
n 
n 
n 
n 
n Bij
ci c j Re[Hi ()H j (−)] = 2ci c j Aij |Hi ()|2 + 2ci c j 2 |Hi ()|2 (16)
i=1 j=1 i=1 j=1 i=1 j=1 i2
j=i j=i j=i

The response power spectral density in (4) is then equal to


⎡ ⎤ ⎡ ⎤
n n n  2 
n
G() = ⎣ci2 +2ci c j Aij ⎦ |Hi ()|2 G o ()+ ⎣2ci 2 c j Bij ⎦ |Hi ()|2 G o () (17)
i=1 j=1 i=1 i j=1
j=i j=i

Copyright 䉷 2010 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2011; 40:1181–1196
DOI: 10.1002/eqe
THE COMPLETE SRSS MODAL COMBINATION RULE 1185

0.3
Definition
0.2

Re [Hi( )Hj ( )]
0.1

-0.1

-0.2

-0.3
0 1 2 3 4 5 6
(a)
0.3
Partial Fractions
0.2
Re [Hi( )Hj ( )]

0.1

-0.1

-0.2

-0.3
0 1 2 3 4 5 6
(b)

Figure 2. Re[Hi ()H j (−)]: (a) from definition (i = 2 rad/s,  j = 4 rad/s, i =  j = 0.02) and (b) from
partial fraction decomposition (i = 2 rad/s,  j = 4 rad/s, i =  j = 0.02).

Let

n
i = ci2 +2ci c j Aij (18)
j=1
j=i


n
i = 2ci c j Bij (19)
j=1
j=i

hence

n 
n 2
G() = i |Hi ()|2 G o ()+ i |Hi ()|2 G o () (20)
i=1 i=1 i2


The kth spectral moment of G(), k = 0 k G() d, is then
 ∞  ∞
n n 
i
k = i k |Hi ()|2 G o () d+ 2+k |Hi ()|2 G o () d (21)
i=1 i 0
2
i=1 0

Consider now the response of unit-mass SDOF modal oscillators with natural frequency i , critical
damping ratio i , subjected to earthquake ground acceleration with power spectrum G o (). The
spectral moments of the modal oscillators responses, k,i , are
 ∞
k,i = k |Hi ()|2 G o () d (22)
0
Thus, in terms of k,i the spectral moments of the response power spectrum can be expressed as

n i
k = i k,i + k+2,i (23)
i=1 i2

Copyright 䉷 2010 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2011; 40:1181–1196
DOI: 10.1002/eqe
1186 E. HEREDIA-ZAVONI

provided the spectral moments k,i and k+2,i exist. As shown in (23) the spectral moments of the
response can be expressed rigorously in terms of the spectral moments of SDOF modal responses
in such a way that fully accounts for modal response variances and cross-covariances. The first
three spectral moments, o , 1 and 2 , are used for characterizing the mean maximum response.
Notice that for some idealized models of G o () not all of the spectral moments k,i and k+2,i
exist; such is the case of ideal white-noise, for which only 0,i , 1,i and 2,i can be computed.
However, for realistic earthquake ground acceleration models of G o (), the spectral moments
k,i and k+2,i necessary to compute o , 1 and 2 shall exist. Some ground motion models for
which spectral moments exists are: (1) band-limited white noise, modelling wide-band ground
motions with smoothly varying spectrum ordinates over a wide-frequency range; (2) the modified
Kanai–Tajimi model, which can represent a variety of power spectrum shapes such as wide-band
spectra for ground motions on firm soil, or narrow band ones for ground motions on soft soil and
(3) harmonic acceleration waves with random phase angle, with power spectrum given by Dirac
delta functions.

Response variance
Using (23), the response variance 2Y = o is


n i
2Y = i 0,i + 2,i (24)
i=1 i2

where 0,i and 2,i are the variances of the displacement and velocity response of the ith modal
oscillator
 ∞
0,i = |Hi ()|2 G o () d (25)
0
 ∞
2,i = 2 |Hi ()|2 G o () d (26)
0

Let, 2di = 0,i , 2vi = 2,i , then from (24)


n i 2
2Y = i 2di + vi (27)
i=1 i2

It is shown in Equation (27) that the structural response variance can be expressed rigorously as
a sum of SDOF modal displacement and velocity response variances, 2di and 2vi , multiplied by
coefficients i and i which depend on modal parameters of the structure. It is noted that expressing
the structural response as a simple sum of uncoupled SDOF modal responses, yet accounting fully
for modal variances and cross-covariances, involves not only the modal displacement variances,
but also the modal velocity variances. Recalling (18) and (19), it is also of interest to notice that the
terms in the right-hand side of (27) are associated with different contributions to the total response:
(1) the term ci2 in i accounts for the contribution from modal response variances; (2) the terms
2ci nj=1, j=i c j Aij in i , and 2ci nj=1, j=i c j Bij in i are both associated with the contribution
from the cross-covariance between modal responses.
The response variance in (27) can also be written as follows:

n  2
2Y = i + i2 2vi 2di (28)
i=1 i di

where vi /di = 2,i is the so-called ‘apparent’ or ‘characteristic’ frequency, which determines
the mean zero-crossing rate of the modal oscillator response in the case of stationary Gaussian

Copyright 䉷 2010 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2011; 40:1181–1196
DOI: 10.1002/eqe
THE COMPLETE SRSS MODAL COMBINATION RULE 1187

excitation. The characteristic frequency 2,i depends on modal frequencies and damping ratios
and on the ground motion input. Let us define a modal response coefficient i
22,i
i = i +i (29)
i2
Then, from (28)

n
2Y = i 2di (30)
i=1

By introducing the characteristic frequency, the response variance in (30) is expressed rigorously in
terms of modal displacement variances only. Both 2di and 2,i can be computed independently for
each mode given the ground acceleration power spectrum G o (). In (30) the response analysis may
proceed on a mode by mode basis thus setting the framework for a response spectrum formulation.
Note that if the cross-covariance contribution is neglected

n 
n
c j Aij = 0, c j Bij = 0,
j=1 j=1
j=i j=i
n
then i = ci2 , i = 0 and i = ci2 , and hence it follows from (30) that 2Y ≈ i=1 ci di ,
2 2 which is the
classical SRSS formulation.

Mean maximum response


From (23), the variance of the response time derivative 2Ẏ = 2 is

n i
2Ẏ = i 2,i + 4,i (31)
i=1 i2
Given that 4,i = 44,i 2di and 2,i = 22,i 2di , where k,i = (k,i /o,i )1/k is the kth characteristic
frequency of a modal oscillator response, then

n i 4
Ẏ =
2
i 2,i + 2 4,i 2di
2
(32)
i=1 i

Let
i = i 22,i +(i /i2 )44,i define a modal response coefficient for the time derivative of the
structural response, then

n
2Ẏ =
i 2di (33)
i=1

Hence, the variance of the response time derivative can also be expressed rigorously as a linear
combination of modal displacement variances 2di . For Gaussian stationary ground acceleration,
the mean maximum response, Ymax = E[max |Y (t)|, 0t ], is

n
Ymax = pY = i p 2 2di (34)
i=1

where peak factor p is given by [4]


 0.5772
p = 2 ln( e )+ √ (35a)
2 ln( e )
⎧ 
⎪ 0.45
⎨  (1.63 −0.38), <0.69

Y
e =  (35b)

⎩ Ẏ , 0.69
Y

Copyright 䉷 2010 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2011; 40:1181–1196
DOI: 10.1002/eqe
1188 E. HEREDIA-ZAVONI


and  = 1−(21 /0 2 ), 01, is a spectral parameter that measures the bandwidth or spread of
G() [6]. For 0.69 the peak factor is equal to that given by Davenport [11].

APPLICATION EXAMPLE

Consider the structural system shown in Figure 3; let m 1 , k1 and m 2 , k2 be the mass and lateral
stiffness of the first and second storey, respectively. Suppose k1 /k2 = m 1 /m 2 = . For  = 1 the
structural system models a two-storey frame with a uniform distribution of lateral stiffness and
mass along height;  = 10 models a frame with rather a large difference in stiffness and mass
between the first and second storeys. The cases  = 100 and 200 model a primary-appendix system.
Four models of ground motion input are considered: ideal and band-limited white noise, and the
modified Kanai–Tajimi power spectra for wide- and narrow-band motions. The c-SRSS formulation
given in Equation (30), in terms of modal displacement variances 2di and apparent frequencies
2,i , is used in this example. For some ground motion models 2di and 2,i are readily available
from analytical solutions; for other ones they are computed numerically.
Let the inter-storey drift be the response of interest. From the analytical solution in Equation
(6) the response variance can be written as follows:

n 
n 
n
2Y = ci2 2di + ci c j ij di dj (36)
i=1 i=1 j=1
j=i

where ij are the cross-correlation coefficients between modal displacements


 ∞
1
ij = Re[Hi ()H j (−)]G o () d (37)
di dj 0
The standard deviation of the relative displacement between second and first storeys was computed
using: (1) the analytical solution given in Equation (36) which we term here as the ‘exact response’;
(2) the c-SRSS rule; (3) the SRSS rule and (4) the original CQC rule [5]. In the original CQC rule
analytical expressions were given for the cross-correlation coefficients ij assuming white-noise
ground motion input; these expressions were proposed for wide-band seismic inputs with arbitrary
power spectrum shapes and modal frequencies that are not much higher than the predominant
frequency of the ground motion [4, 12]. The advantage of the CQC formulation is that expressing ij
in terms of only modal frequencies and damping ratios, allows the spectral moments of the response
to be given in terms of spectral moments of SDOF modal responses, which can subsequently
be defined in terms of mean response spectrum ordinates. For narrow-band input motions, for
instance ground accelerations on soft soil deposits recorded in Mexico City during the September,

m2

k2
m1

k1

Figure 3. Structural system.

Copyright 䉷 2010 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2011; 40:1181–1196
DOI: 10.1002/eqe
THE COMPLETE SRSS MODAL COMBINATION RULE 1189

Table I. Response comparison for white-noise ground acceleration.


1 2 Y /G o Y /G o Y /G o Y /G o
1 Y (SRSS)
 (rad/s) (rad/s) 2 ‘Exact’ (c-SRSS) (SRSS) (CQC) Y (c-SRSS)

1 4.94 12.94 0.38 0.234 0.234 0.234 0.234 1.00


10 6.83 9.36 0.73 0.554 0.554 0.577 0.554 1.04
100 7.61 8.41 0.90 1.251 1.251 1.759 1.251 1.40
200 7.72 8.28 0.93 1.434 1.434 2.484 1.434 1.73

1985 earthquake, the approximation of ij based on white-noise inputs may not be appropriate, as
has already been pointed out by Der Kiureghian and Nakamura [12]. Of course, for narrow-band
motions one can compute the exact values of ij considering the frequencies and damping ratios
of modes ‘i’ and ‘ j’. In [12], the exact values of ij were computed for narrow-band motions and
then used to estimate the mean maximum response. Notice however that a combination rule or
response spectrum formulation where the values of ij need to be computed for all possible cross-
combinations of modal frequencies i ,  j , and cross-combinations of modal damping ratios i ,  j ,
differs in essence from the philosophy of a rule that aims at providing a practical way to compute
the response on a mode by mode basis facilitating the use of response spectra. Furthermore, having
to compute the exact values of ij as given in Equation (37) is the same task involved in computing
the ‘exact’ response using the random vibration general solution given in Equation (36). The CQC
rule as originally proposed in [5] to replace the SRSS rule for engineering practice is thus used
here for comparison purposes, although it is recognized that its approximation of ij may not be
appropriate for narrow-band motions.
The modal frequencies and the √ standard deviation of the response to an ideal white-noise ground
acceleration considering o = k2 /m 2 = 8 rad/s and 5% modal damping are listed in Table I.
For ideal white-noise input G o () = G o , 2,i = i and thus i = i +i depends on structural
parameters only. As  increases from 1 to 200, modal frequencies 1 , 2 , become closely spaced
and 1 /2 → 1. Table I shows that for all cases the c-SRSS rule yields the same results as the
exact solution, thus verifying that the formulation is correct. The CQC rule also yields the same
results as the exact solution for all cases, showing that for wide-band inputs it accounts accurately
for modal response cross-covariances as well. The SRSS rule estimates the response correctly for
 = 1 and with small error for  = 10. However, for the primary-appendix system, where modal
frequencies are closely spaced (1 /2 = 0.90 and 0.93), the SRSS estimate is 40 and 73% greater
than the exact solution for  = 100 and 200, respectively. Since the ground motion is a wide-band
process, the error using SRSS depends only on the separation between modal frequencies. These
results show that in contrast to SRSS, the c-SRSS is capable of taking into account the contributions
from modal cross-covariance and yield the correct response as modal frequencies become closely
spaced.
Consider the case of band-limited white-noise ground acceleration. For band-limited white noise,
G o () = G o , a , G o () = 0, >a , which has been used to model ground acceleration on
firm soils, a closed-form solution also exists for 2,i although it is not as simple as for ideal
white-noise. Figure 4 shows the variation of 2,i /i versus a /i for band-limited white-noise
excitation for 2 and 5% critical damping ratios. Notice that the values of 2,i /i for the band-
limited white-noise case will approach the ideal white-noise case as a /i increases. For instance,
for a /i 2, the difference in 2,i /i between band-limited and ideal white-noise is already
less than 2%. The response standard deviation is listed in Table II considering a = 45rad/s;
characteristic frequencies of the two modal responses, 2,1 and 2,2 are also given in Table II.
The c-SRSS and CQC results are the same, and additionally are equal to those for white-noise
excitation. This is so because a /i is at least greater than 3.5, and thus the results are, at least
to three decimal digits, equal to those for ideal white-noise. For  = 1 and 10, the SRSS yields
reasonable estimates of the response. For the primary-appendix system, due to closely spaced

Copyright 䉷 2010 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2011; 40:1181–1196
DOI: 10.1002/eqe
1190 E. HEREDIA-ZAVONI

0.98

2i / ω i
0.96

ξi = 2%
2%
ξi = 5%
5%

0.94
1 1.2 1.4 1.6 1.8 2
ω a/ ωi

Figure 4. Variation of 2,i /i versus a /i for response to band-limited white noise.

Table II. Response comparison for band-limited white-noise ground acceleration.


Y (SRSS)
 2,1 (rad/s) 2,2 (rad/s) Y /G o (c-SRSS) Y /G o (SRSS) Y /G o (CQC) Y (c-SRSS)

1 4.92 12.81 0.234 0.234 0.234 1.00


10 6.79 9.29 0.554 0.577 0.553 1.04
100 7.56 8.35 1.251 1.759 1.250 1.40
200 7.67 8.23 1.434 2.484 1.431 1.73

Table III. Response comparison for wide-band ground acceleration.


Y (SRSS)
 Y /G o ‘Exact’ Y /G o (c-SRSS) Y /G o SRSS Y /G o (CQC) Y (c-SRSS)

1 0.265 0.265 0.264 0.264 1.00


10 0.679 0.679 0.700 0.671 1.03
100 1.556 1.556 2.163 1.536 1.39
200 1.785 1.785 3.057 1.761 1.71

modal frequencies, the SRSS rule overestimates the response by 40–70%. Notice that frequencies
2,i are very close to modal frequencies i .
Response results were obtained for ground accelerations modelled by the modified Kanai–Tajimi
power spectrum [13]

1+42f (/ f )2 G r (/g )4


G o () = (38)
[1−(/ f )2 ]2 +42f (/ f )2 [1−(/g )2 ]2 +42g (/g )2

where  f , g ,  f , g can be thought of as characteristic-ground frequencies and damping, respec-


tively, and G r is a white-noise amplitude at rock basement. Table III shows results for wide-band
ground accelerations using  f = 15 rad/s, g = 0.5 rad/s and  f = g = 0.6 [8]. The c-SRSS rule
yields the exact response for all cases; the CQC rule estimates the response correctly. The SRSS
rule overestimates the response by 39 and 71% for  = 100 and 200, respectively; the SRSS yields
the correct response for  = 1 and very little error for  = 10.
The response to a narrow-band seismic input modelled by the modified Kanai–Tajimi power
spectrum for  f = , g = 0.5 rad/s and  f = 0.2,g = 0.6, is listed in Table IV. The c-SRSS rule
yields the same response as the exact solution for all cases. The SRSS rule estimates with little
error the response only for  = 1; the SRSS rule overestimated the response introducing gross errors

Copyright 䉷 2010 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2011; 40:1181–1196
DOI: 10.1002/eqe
THE COMPLETE SRSS MODAL COMBINATION RULE 1191

Table IV. Response analyses for narrow-band ground acceleration.


Y (SRSS) Y (CQC)
f  Y /G o ‘Exact’ Y /G o (c-SRSS) Y /G o (SRSS) Y /G o (CQC) Y (c-SRSS) Y (c-SRSS)

3.14 1 0.211 0.211 0.221 0.221 1.05 1.05


10 0.208 0.208 0.303 0.292 1.45 1.40
100 0.353 0.353 0.832 0.598 2.36 1.70
200 0.395 0.395 1.169 0.682 2.95 1.72

Table V. Modal response variance and cross-covariance (narrow-band ground acceleration).


1 2
f  f f 1,2 ‘Analytical’ 1,2 (CQC) 21 22

3.14 1 1.57 4.12 0.603 0.008 0.2435 0.0013


10 2.17 2.98 0.715 0.089 0.0316 0.0060
100 2.42 2.68 0.848 0.499 0.0174 0.0103
200 2.46 2.64 0.900 0.670 0.0161 0.0112

Table VI. Response analyses for narrow-band ground motion inputs.


Y (SRSS) Y (CQC)
f  Y /G o ‘Exact’ Y /G o (c-SRSS) Y /G o (SRSS) Y /G o (CQC) Y (c-SRSS) Y (c-SRSS)

5 1 0.562 0.562 0.564 0.564 1.00 1.00


10 0.580 0.580 0.679 0.661 1.17 1.13
100 0.971 0.969 1.684 1.225 1.73 1.26
200 1.089 1.088 2.349 1.388 2.16 1.27
8 1 0.364 0.364 0.358 0.357 0.98 0.98
10 1.292 1.292 1.264 1.216 0.98 0.94
100 3.269 3.269 4.254 3.042 1.30 0.93
200 3.777 3.776 6.052 3.511 1.60 0.93

from 45 to 195% for the systems with close modal frequencies. Similarly, the CQC rule produced
acceptable results for  = 1 only; for the other  values the response is overestimated by CQC and
significant errors of the order of 40–70% are obtained. Der Kiureghian and Nakamura [12] noted
that for narrow-band inputs significant errors are obtained in the computation of ij using the CQC
white-noise formulation when modal frequencies are much higher than the predominant frequency
of the input ground motion, i.e. when i >> f . Table V lists the ratio of modal frequencies to
predominant frequency of the input ground motion, i / f , the modal response variances, 21 , 22
and the cross-correlation coefficient 1,2 between modal responses. For  =1, the cross-correlation
coefficient between first and second mode responses is 1,2 = 0.603; using CQC yields 1,2 = 0.008.
CQC does not estimate 1,2 correctly, however its influence on the computation of the response
is negligible. This is due to the fact that for  = 1, 21 = 0.2435 and 22 = 0.0013; thus, the cross-
covariance between modal responses is much less than the first mode variance, which is dominant
in the computation of the response. Consequently, even though the CQC formula does not estimate
1,2 accurately, this error has very little effect on the computation of the response. On the other
hand, when  = 200, 12 = 0.9; using CQC 12 = 0.67, which is much more accurate in this case;
however, now 21 = 0.0161, 22 = 0.0112, and the contribution of the modal cross-covariance is of
the same order of magnitude as the modal variances. Hence, in this case the error in the estimation
of ij has a greater effect on the estimation of the response, as shown by the results listed in
Table IV.
Additional results are shown in Table VI for other values of ratio i / f considering  f =
5 rad/s and 8 rad/s. It can be seen that in all cases the c-SRSS rule yields the ‘exact’ response.
Using SRSS errors in the response estimates are 30 and 60% ( f = 8 rad/s), and 73 and 116%

Copyright 䉷 2010 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2011; 40:1181–1196
DOI: 10.1002/eqe
1192 E. HEREDIA-ZAVONI

Table VII. Modal response variance and cross-covariance (narrow-band ground motion inputs).
1 2
f  f f 1,2 ‘Analytical’ 1,2 (CQC) 21 22

5 1 0.99 2.59 0.1018 0.0088 1.5906 0.0030


10 1.37 1.87 0.4498 0.0896 0.1705 0.0189
100 1.52 1.68 0.7085 0.4993 0.0759 0.0378
200 1.54 1.66 0.8083 0.6704 0.0685 0.0420
8 1 0.62 1.62 −0.1275 0.0088 0.6287 0.0120
10 0.85 1.17 −0.0551 0.0896 0.5159 0.1394
100 0.95 1.05 0.4183 0.4993 0.4375 0.2883
200 0.97 1.04 0.6168 0.6704 0.4197 0.3138

( f = 5 rad/s), for  = 100 and 200, respectively. Error in the estimation of the response using
CQC is less than 7% for  f = 8 rad/s and can be greater than 25% for  f = 5 rad/s. The values of
i / f , 1,2 , 21 and 22 corresponding to these cases are shown in Table VII. As 1 , 2 become
greater than  f a larger error is introduced in the computation of 1,2 using CQC. For instance,
for  f = 5 rad/s, 1,2 = 0.7085 and 0.8083 for  = 100 and 200, respectively, whereas using CQC
yields 1,2 = 0.4993 and 0.6704. When  f = 8 rad/s and 1 and 2 are about 0.6 f –1.05 f ,
1,2 is estimated more accurately. As stated above, the accuracy of the response estimate depends
also on the relative importance of modal cross-covariance terms with respect to modal variances.
Thus, for  = 1, although CQC does not estimate ij correctly, no significant errors are introduced
in the response because the modal cross-covariance is negligible.

RESPONSE SPECTRUM FORMULATION BASED ON THE C-SRSS RULE

For practical applications, a response spectrum formulation can be derived from the c-SRSS rule.
Let Sdi = Sd (i , i ) denote the mean maximum relative displacement of a SDOF system with
natural frequency i and critical damping ratio i subjected to a stationary ground acceleration;
then Sdi = pdi di , where pdi is the corresponding peak factor. It follows from (34) that:
  2
n p
Ymax = i Sdi (39)
i=1 pdi
It is reasonable and widely accepted to assume that p/ pdi ≈ 1, as has been done in deriving
other response spectrum combination rules. Thus, under this assumption the maximum structural
response can be estimated from


n
Ymax ≈ i (Sdi )2 (40)
i=1

Note that when the cross-covariance between modal responses is neglected, then i = ci2 , i = 0
and i = ci2 ; hence


n
Ymax ≈ (ci Sdi )2 (41)
i=1

which is the classical response spectrum formulation based on Rosenblueth’s SRSS rule. In (40)
both Sd,i = Sd (i , i ) and the characteristic frequency 2,i , involved in the computation of i , need
to be given. The characteristic frequency can be computed from a power spectral density consistent
with a given response spectrum. As with other response spectrum-based modal combination rules
published in the literature, it is necessary to be able to specify spectrum-consistent power spectral
density (PSD) since these rules generally follow from random vibration formulations where seismic

Copyright 䉷 2010 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2011; 40:1181–1196
DOI: 10.1002/eqe
THE COMPLETE SRSS MODAL COMBINATION RULE 1193

200
150
100
50

a (cm/s )
2
0
-50
-100
-150
-200
0 20 40 60 80 100 120 140 160
t (s)

Figure 5. SCT (EW) record; Mexico earthquake of September 19, 1985 (Ms = 8.1).

ground motion is characterized by PSDs rather than by response spectra. Several methods have
been developed for generating response-spectrum-consistent PSD and higher order approximations
of the power spectrum can be obtained (see e.g. Der Kiureghian and Nakamura [12], Vanmarcke
[7] and Der Kiureghian and Nueunhofer [8]). However, a consistent PSD might represent the
nature of the ground motion in an approximate way, for instance regarding non-stationarity. Thus,
spectral parameters such as the characteristic frequency, computed based on a consistent PSD, may
involve some degrees of approximation in representing the true nature of ground motion.
It has been shown in the application example above that for narrow-band ground motions the
c-SRSS rule is able to yield the correct response, which is not always the case for the SRSS or for
the original CQC rules. It is of interest now to examine the estimation of the structural response
using the response spectrum formulation for narrow-band inputs. The SCT (EW) record of the
Mexico City earthquake of September 19th, 1985 (Ms = 8.1) was selected to model a narrow-band
ground motion input. The relative displacement between second and first storeys of the structural
system in Figure 3 was estimated using the c-SRSS response spectrum formulation in (40). The
c-SRSS response estimates were compared to the exact response obtained from performing time-
domain analyses using simulations of the SCT (EW) ground acceleration record. The response was
also estimated using the SRSS and CQC response spectra formulations. A modified Kanai–Tajimi
spectral density function was used with parameters  f = 3.10 rad/s, g = 2.29 rad/s,  f = 2.2%,
g = 4.9% which have been estimated by nonlinear regression analyses based on the SCT (EW)
ground acceleration record [14]. An ensemble of 50 ground motion realizations were simulated
using Yeh and Wen method with appropriate envelope and frequency modulation functions [15].
Figure 5 shows the SCT (EW) record. Figure 6 shows the modified Kanai–Tajimi model, and
Figure 7 shows the mean displacement response spectra for the 50 ground motion realizations.
Table VIII lists values of the mean modal spectral displacements, Sdi , the modal characteristic
frequencies computed by numerical integration of the ground acceleration PSD, and the modal and
total response peak factors. It is seen that peak factor ratios p/ pdi are less than 1.03. Table IX lists
the mean maximum response obtained from the time-domain analyses and the response spectrum
formulations. For well-separated modes ( = 1) the c-SRSS response spectrum rule yields an
estimate of the maximum response that, for practical purposes, is equal to the exact response. The
SRSS and CQC rules yield reasonably accurate estimates of the response with errors of the order
of 10%. For closely spaced modal frequencies ( = 100 and 200), the c-SRSS response spectrum
rule yields good estimates that are within 10% of the exact response. The SRSS estimates are 5–7
times as large as the exact response. Using CQC overestimates the response by a factor of 4.
Suppose that the modal characteristic frequencies were not computed by numerical integration
and were approximated instead using one of the following criteria: (1) the dominant frequency in
the power spectral density (PSD) consistent with the mean response displacement spectrum and (2)
the dominant frequency in the PSD consistent with the response displacement spectrum for the SCT
record. The PSD consistent with given response spectrum were computed following the first order
approximation applied by Der Kiureghian and Nakamura [12]. The dominant frequencies were

Copyright 䉷 2010 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2011; 40:1181–1196
DOI: 10.1002/eqe
1194 E. HEREDIA-ZAVONI

3000

2500

2000

Go (ω)/Gr
1500

1000

500

0
0 1 2 3 4 5
ω (rad/s)

Figure 6. Kanai–Tajimi spectral density model for soft soil.

140

120

100
Sd (cm)

80

60

40

20

0
0 1 2 3 4 5 6 7
T (s)

Figure 7. Mean response spectrum.

Table VIII. Mean spectral displacements and response parameters.


 Sd1 (cm) Sd2 (cm) 1 2 pd1 pd2 p
1 12.04 1.13 3.06 2.97 2.10 2.09 2.10
100 3.68 2.92 2.99 2.99 2.09 2.09 2.14
200 3.55 3.02 2.99 2.99 2.09 2.09 2.16

Table IX. Comparison of response using response spectra formulations.


Relative displacement (cm)
Time-domain c-SRSS (PSD—mean c-SRSS (PSD—SCT
 analysis c-SRSS SRSS CQC response spectrum) response spectrum)
1 4.90 4.87 5.40 5.40 4.87 4.87
100 4.29 4.28 23.45 16.81 4.76 4.89
200 4.26 4.69 32.93 19.15 5.26 5.42

taken as those corresponding to the peak value of the PSD. The dominant frequencies obtained for
each of the two criteria were: (1) 3.06 rad/s and (2) 3.08 rad/s, respectively. Table IX includes the
response estimates using these choices of characteristic frequencies. It is seen that for the system
with well-separated modal frequencies modes ( = 1), the choice of characteristic frequency has
no effect on the response estimate. For the systems with  = 100 and 200, the response estimates

Copyright 䉷 2010 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2011; 40:1181–1196
DOI: 10.1002/eqe
THE COMPLETE SRSS MODAL COMBINATION RULE 1195

using these criteria are about 15 and 25% greater than the exact response, respectively, giving an
indication of the sensitivity of the response estimates. It is relevant to notice that in any case these
estimates are in much closer agreement with the exact response than the SRSS and CQC estimates
which as indicated before overestimate the response by factors greater than 4.

CONCLUSIONS

The complete Square-Root-of-Sum-of-Squares (c-SRSS) modal combination rule has been derived.
In the c-SRSS rule the structural response is expressed rigorously in terms of uncoupled SDOF
modal oscillator responses in such a way that fully accounts for the contributions from both modal
response variances and cross-covariances. It is an improvement over the classical SRSS rule that
only accounts for modal response variances. The c-SRSS rule expresses the spectral moments of
the structural response in terms of the moments of uncoupled SDOF modal oscillator responses.
Therefore, the analysis of the structural system can be reduced to that of a series of SDOF oscillators
with natural frequencies and damping ratios equal to the modal frequencies and damping ratios of
the structure. In this formulation there is no need to compute cross-correlation coefficients between
modal responses. All the parameters in the c-SRSS rule depend either on structural properties or
on spectral parameters of SDOF modal oscillator responses. It has been shown that expressing the
structural response as a simple sum of uncoupled SDOF modal responses, yet accounting fully for
modal variances and cross-covariances, involves not only the modal displacement variances, but
also the modal velocity variances. Introducing the modal characteristic frequencies, the response
variance in the c-SRSS rule is expressed as a simple sum of modal displacement variances,
multiplied by response coefficients which can be computed in a straightforward manner given the
modal frequencies and damping ratios. The c-SRSS rule reduces to the classical SRSS rule when
modal cross-covariances are neglected. Considering that in the c-SRSS rule the response analysis
proceeds on the basis of summing uncoupled SDOF modal responses, and that a mean response
spectrum can be specified in terms of the ground acceleration power spectral density, the c-SRSS
rule provides the basis for developing a response spectrum formulation for MDOF systems.
By means of an example, it was verified that the c-SRSS rule yields the correct response
for systems with either separated or closely spaced modal frequencies and for seismic inputs
with arbitrary power spectrum, from wide-band to narrow-band ground motions. The results were
compared against the SRSS. As known, the SRSS introduced large errors in the computation of
the response of systems with closely spaced modal frequencies. Results using the c-SRSS rule
have also shown that for structural systems subjected to narrow-band ground motions, such as
those characteristic of soft soil deposits, the contribution from the cross-covariance between modal
responses is not necessarily small and should not always be neglected in the analysis. Results were
also compared against the CQC rule. For wide-band seismic inputs the c-SRSS and CQC rules
yielded the same results as the exact response. However, for narrow-band motions it is known
that the CQC rule cannot estimate the response correctly and its accuracy depends on the relation
between modal frequencies and the ground motion dominant frequency. On the other hand, the
c-SRSS rule has been proven to yield the correct response for both wide- and narrow-band input
motion.
Based on the c-SRSS rule, a response spectrum formulation was introduced for practical appli-
cations to estimate the maximum response. It was applied to an example considering narrow-band
input motion based on the Mexico City SCT(EW) ground acceleration record from the great
earthquake of September 19, 1985. The results showed that the c-SRSS response spectrum formu-
lation yields good estimates of the response; for systems with closely spaced modal frequencies,
it provides more accurate results than the formulations based on other combination rules. Results
also showed the sensitivity of the proposed formulation to the estimated values of the character-
istic frequencies; it was noted that even when using estimates of the characteristic frequencies the
response obtained was in much closer agreement with the exact response than when using other
rules. However, further tests need to be carried out to understand the sensitivity of the proposed

Copyright 䉷 2010 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2011; 40:1181–1196
DOI: 10.1002/eqe
1196 E. HEREDIA-ZAVONI

formulation to errors in the estimation of the characteristic frequency. It is finally noted that
since the response variance depends on both modal displacement and modal velocity variances, as
given in Equation (27), future studies may consider proposing a response formulation in terms of
displacement and velocity response spectra. For such possible formulation it should be assessed
whether the approximation p/ pdi ≈ 1, which holds for displacement peak factors, is also reasonable
for the velocity peak factors in some modal frequency ranges. If it was not the case it would then
be necessary to assess the accuracy of a formulation based on such an assumption in comparison
to the response spectrum formulation that has been presented here and to other rules.

REFERENCES
1. Rosenblueth E. A basis for aseismic design. Ph.D. Thesis, University of Illinois, Urbana, 1951.
2. Rosenblueth E, Elorduy J. Responses of linear systems to certain transient disturbances. Proceedings of the
Fourth World Conference on Earthquake Engineering, Santiago de Chile, 1969; A1-185.
3. Der Kiureghian A. Structural response to stationary excitation. Journal of the Engineering Mechanics Division
(ASCE) 1980; 106(EM6):1195–1213.
4. Der Kiureghian A. A response spectrum method for random vibration analysis of MDF systems. Earthquake
Engineering and Structural Dynamics 1981; 9:419–435.
5. Wilson EL, Der Kiureghian A, Bayo EP. A replacement for the SRSS method in seismic analysis. Earthquake
Engineering and Structural Dynamics 1981; 9:187–194.
6. Vanmarcke EH. Properties of spectral moments with applications to random vibration. Journal of the Engineering
Mechanics Division (ASCE) 1972; 98(EM-2):425–446.
7. Vanmarcke EH. Structural response to earthquakes. In Seismic Risk and Engineering Decisions Developments in
Geotechnical Engineering, Chapter 8, Lomnitz C, Rosenblueth E (eds), vol. 15. Elsevier: Amsterdam, 1976.
8. Der Kiureghiuan A, Neuenhofer A. A response spectrum method for multi-support seismic excitations. Report
No. EERC 91/08, University of California, Berkeley, 1991.
9. Heredia-Zavoni E, Vanmarcke EH. Seismic random vibration analysis of multi-support structural systems. Journal
of Engineering Mechanics (ASCE) 1994; 120:1107–1128.
10. Heredia-Zavoni E, Vanmarcke EH. Random-vibration-based response spectrum method for multi-support structural
systems. Journal of Seismology and Earthquake Engineering 1998; 1(1):35–50.
11. Davenport AG. Note on the distribution of the largest value of a random function with applications to gust
loading. Proceedings of the Institution of Civil Engineers, London, 1964; 28:187–196.
12. Der Kiureghian A, Nakamura Y. CQC modal combination rule for high-frequency modes. Earthquake Engineering
and Structural Dynamics 1993; 22:943–956.
13. Clough RW, Penzien J. Dynamics of Structures. McGraw-Hill: New York, 1975.
14. Silva F. Calibración del método de linealización equivalente estática para sistemas histeréticos simples. M. Eng.
Thesis, Facultad de Ingeniería, DEPFI, Universidad Nacional Autónoma de México, 1998.
15. Yeh CH, Wen YK. Modeling of nonstationary earthquake ground motion and biaxial and torsional responses of
inelastic structures. Civil Engineering Studies, Structural Research Series No. 546, University of Illinois, Urbana,
1989.

Copyright 䉷 2010 John Wiley & Sons, Ltd. Earthquake Engng Struct. Dyn. 2011; 40:1181–1196
DOI: 10.1002/eqe

You might also like