At'.

A;~~~~~~ ' ,' ~ ._ jNf

TRANSACTIONS.j
of the
6th International Conference on

STRUCTURAL LWSt=Rt
MECHANICS 00 :
INREACTOR1
TECHNOLOGY
Palais des Congres
i4
Paris, France a 17-21 August 1981

Vol. K (a). Seismic Response Analysisof,

Nuclear Power Plant Systlemrs

,_'

Conference organization by:

International Association for Structural Mechanics in Reactor Technology
Commission of the European Communities, Brussels
SFEN - Soci6t6 Francaise d'Energie Nucl6aire, Paris

NORTH-HOLLAND PUBLISHING COMPANY

for
THE COMMISSION OF THE EUROPEAN COMMUNITIES
 K 7/5
Combination of Mlodal Responses

a'3-. A.K. Gupta, K. Cordero

North Carolina State University, Departmnent of Civil Engineering,
Box 5993, Raleigh. North Carolina 27650. U.SA.

In the response spectrum method of analyzing structures under seismic loads, vilues
of the maximum probable responses are calculated in each mode of vibration. These :epne
can be combined using a double sum equation if the correlation between the modal res3ponses
ia known. An approximate expression for the correlation coefficient given by Rosenlhlueth
and Elorduy[l] does not apply in the low and high frequency ranges. It is shown thaitthe
modal responses can be assumed to consist of two components, a rigid component, and a
damped periodic component. Various modal rigid components are perfectly correlated. An
expression for the co~rrelation between the damped periodic components is presented which
is similar to that presented by Rosenblueth and Elorduy~l].' An equation is derived for the
correlation between modal responses on this basis, which is shown to be in good agrnement
'with the numerically obtained values.

IjAH
 Introduction There are too other potential pr
1.
In the response spectrum method of analyzing structures under seismic loads, the values modal frequencies are higher than the

of the maximum probable responses are calculated in each mode of vibration. The variation In fact, the response time histories
ii
of responses with time is not known, therefore, there is no "exact" way of combining the would be almost ,erfectly correlated,

maximum modal responses. If the maximum response in mode i is Ri, it can be shown that the sufficiently apart. The writers foun

combined maximum response, R, would satisfy the following inequality than lHz, significant correlation bet
discussed further in the subsequent E
EIR I >R> E (1) 2
Xennedy[ ] is the following. When tt
the left hand side being an upper bound, and the right hand side a lower bound. In many certain point, the correlation betwef
cases, when the modal frequencies are sufficiently far apart, the right hand side, commonly than decrease as predicted by Eq(3).
known as the SRSS (square root of the sum of the squares) combination, gives reasonably quite likely that the high frequency
accurate values. Statistically, in these cases, the modal responses can be considered to when the low frequency response read
be independent, or uncorrelated. When the modal frequencies are relatively close, however, present study, lowever, no such tren
the modal responses no longer remain independent, and then the combined response can be different segmerts of the ground mot
significantly greater than the SRSS value. In fact, when two of the modes have identical unlikely that th.esame segment of th
frequencies, it stands reasons that their responses should be directly added in which cases frequencies. Thus, the second poten
the two responses are perfectly correlated. degree.
In general, for combining responses from two modes one can write A practica. method of calculati
2 2 2 R 2 applicability, :.spresented in the f
R R + R 2 + 22 RR. (2)
method is based, is presented in the
in which
2. Proposed Method of Calculating
l~c 12> r t. :

:: J'J,
Based on Cie observation of the
which depends on the correlation between the modal responses. Theoretically, one could

I
assumption is mide: Any modal respc
think of a negative correlation also, or when c12 < 0, but as will be seen later, it is not
response, Rp, waich has characterist
very likely. Assuming that the earthquake motion can be represented by a finite duration
of the white noise[l], and a rigid
segment of the white noise, Rosenblueth and Elorduy[l] have obtained an approximate equation
input ground motion. It is further
similar to Eq(2). Based on their work, the value of £12 can be written as
R - Ri + Ri2

e12_ +;(f + f1 ) + 2/(td) ' (3) Thus we can write

I
w {l 2
where Rr a
fl - molal frequencies, Hz and
C = critical damping ratio RP= Fl-= Ri
td = duration of the earthquake motion, secs
When two modal responses R1 and R2
It can be seen from Eq(3) that c12-1 for f2 - fl and c12 0 for f2 f1 . Although, Eq(3)
combines response is given by
is approximate it works well for a range of frequencies. Some of the problems are discussed
2 Rr 2 + Rp 2
below.
One of the problems with Eq(3) is that it is not clear as to what value of td should in which

be used. When the frequecies, f, and f2, are sufficiently large, and for relatively large Rr R1 + m2 R2 (perfectly
p2 p2+p 2+P PP
values of the critical damping ratio, the term consisting of td does not play a significant R - R1 + R2 2+ 12R1R2
role. In those :ases, it does not matter what value of td is used in the equation as long
In Eq(7) c.12ii the correlation coe
as it is reasonable. However, when the frequencies are small, the 2/(ltd) term increases
and is defined by an equation simi]
the effective vaLue of the damping, as intended, thus giving a larger value of the c12. 1 f+ f
t
It is in this case the value of c12 is quite sensitive to the value of d' Using the 12 1+ f2 + d1 2
complete duratio:i of the ground motion for td does not appear to be rational.

:1
jij:

-2- K 7
I
 There are two other potential problems pointed out by Kennedyf2]. First, wher, the

inder seismic loads, the values modal frequencies are higher than the maximum ground motion frequency, Eq(3), does not hold.
of vibration. The variation In fact, the response time histories will be practically scaled input time histories, and

Kact" way of combining tie would be almost perfectly correlated, in which case c12 - 1.0, even when f, and f2 are
3 Ri. it can be shown that the sufficiently apart. The writers found that even at other frequencies in the range greater
;uality than lHz, significant correlation between modes existed. This particular aspect ie
discussed further in the subsequent sections. The other potential problem pointed by
(1)
Yennedy[2] is the following. When the modal frequencies are sufficiently apart, beyond a
ide a lower bound. In many certain point, the correlation between the modal responses may start increasing, rather
the right hand side, commonly than decrease as predicted by Eq(3). Heuristically, the reason is simply that it %ould be
,ination, gives reasonably quite likely that the high frequency response, can easily be maximum about the same time
sponses can be considered to when the low frequency response reaches the maximum. In the numerical work done fcr the
are relatively close, however, present study, however, no such trend was evident. The explanation for that is perhaps that
a combined response can be different segments of the ground motion have different frequency contents. As such, it is
of the modes have identical unlikely that the same segment of the motion would excite two modes with widely disparate
directly added in which cases frequencies. Thus, the second potential problem was not found to exist to any discernable
degree.
an write A; A practical method of calculating the correlation factor E12, which has a wide range of
(2) applicability, is presented in the following section. The numerical work on which this
method is based, is presented in the subsequent section.
I 2. Proposed Method of Cdlculating tfii

Theoretically, one could
will be seen later, it
esented by a finite duration
btained an approximate equation
be written as
(3)
Based on the observation of the modal responses and their combinations, a heuristic
assumption is made: Any modal response Ri consists of two parts, a damped periodic
is not
response, RP, which has characteristics similar to that obtained by using a finite segment
4i
of the white noise[l], and a rigid response, Ri, which is perfectly correlated with the
input ground motion. It is further assumed that the two parts are mutually uncorrelated, ie.
2
R Rp2 + Rr (4)
i i i
Thus we can write

and i Ri

tw i i i 1P 2 with frqenis n
.-When two modal responses R, and R and f2 are to be combined, then the
frequencies f1with
for f2 » f1. Although, Eq(3)
combines response is given by
e of the problems are discussed
2 2
R -R +RP (6)
s to what value of td should in which
rge, and for relatively large Rr ' a R + a R (perfectly correlated)
I11 2 2
2
td does not play a significant Rp P -R1 pI Rp 2 + 2cp12RIR2
p+ 2~ p p(7 (7)
used in the equation as long
In Eq(7) cp is the correlation coefficient for the damped periodic part of the responses,
the 2/(Rftd) term increases 12
and is defined by an equation similar to Eq(3)
a larger value of the c12
Cp +
U f
2
f
1
value of td. Using the
to be rational.
12 ~(f2 + f ) +CII(8
 The intent o:~ the present study i
where
2
-f 1 ) > ° (9) in conjunction with the response spect
C1 2 (1-30) (.036 - If2
here replaces, are commonly used in design, such as I
when Eq(9) gives a negative c value it is taken to be zero. The term c12
12
12 1ti ground motions. Similarly, the preser
the value of td to be used with Eq(3) varies
the 2/(std) term in Eq(3). In effect, thus _ 2 d . a histories, which are listed below
2 f1 I. Equations (2), (4) - (7) yield
with the amount of critical damping and with If2
Parkfield, 1966, N50E and N40W
102++12= 2 p
(10) El Centro, 1940, NS and EW
12 (12__al ) (1 a2 ) E1 2
El Centro, :.934, NS and EW
c12 which are quite close to the numerically
It was found that Eq(10) gives values of
including high frequencies. Taft, 1952, N21E and S69E
calculated values for a wide range of frequencies Olympia, 1949. S04E and S86W
* .. .Y
' -. n .- A PrnN i-,,
VAl,.,V
'2 ' 2 I' 12 -u n ~v For any pair of frequencies f, and f2
calculated using the responses from t
£12 =01
When f2is infinite, R2 (t) is ec

of a implicit in the present study. The frequency fl, the factor all is then
which is actually the mathematical definition
,:
and are also a function of the critical damping c were calculated for various freque,
values of a vary with the modal frequency,
to :-i;
to evaluate the values of a, which was found agreement with fq(12).
ratio. The following equation can be used
agreement with the numerically calculated values Fig 1 shows. the comparison of E.
be in reasonable
(12)
(a + 0.1) 'a - m &n f + a) - b, -0.1 a(l.0 from the average of the values calcu
taken equal to 1.0, r
greater than 1, it should be critical damping ratio of 1M. In Eq
If the above equation gives a value of a
set equal to -0.1. ' .Il
than -0.1, it should be As shown, the agreement between the
and similarly, :.fthe equation gives a less
of the constant m, a and b vary with the damping ratio and are given by Theagreements fr other damping rati
The values
made
17 34) effect verifies the assumptions
m - 0.0737:1 tn(-
components: rigid and damped perioc
a - -0.3437 tn (7.594 () f is small, it is seen in Fig 2 th,
(13)
b - -. 03237 tn (14.28 4) coefficients hive a small negative
3. Numerican Tesults values can be Attributed to a limit,
and R2(t) of two single degree of freedom
Consider tle response time histories R1(t) time histories. In any case, since
damping ratio C, subjected to the same earthquake
systems, frequerceis fl and f2 , critical not attach much significance to the
as
and a2' and the covariance a12 are defined
ground motion. The standard deviations, a0 Next, the effect of any error
follows: the value of tie combined response
al _ 1t d - 2 (t)dt of the correlation coefficents, whe
td 2
0 1d when they are also equal in magnitt
2 1 td -22
02 =
d
~ 2 (t)dt R2 _ 2R12 x(1 +t 1 2
) R1

If a different value of the correl;

a02 It I d 1(t) R2(t) dt (14) combined respcnse, R', would be ob
combined respconse is given by
The correlation between the two responses is given by [3]
2 (l 2 _(l-) x 100
a12
010y2 (15) There are two types of variations
-12
value of c12 represented by the at
It is noted that whereas all a2 and 012 are quite sensitive to the value of the duration 77;;
Eq(15) is practically independent of the because of using the idealized eqt
v of ground motion, td' the value of c12 given by
of the responses are covered in the assumed calculate the mean value of £12.
value of td' as long as all significant part
therefore, is not pursued any further here and the effect of dispersion in t1
duratibn. The s:atement can be easily verified,
error in the response value by sul

AA1
-4- - K
 The intent of the present study is to arrive at the values of £12 which can le used
(9) in conjunction with the response spectrum method of analysis. The response spectra which
are commonly used in design, such as in ref[4] are based on a number of measured earthquake
*. The term c1 2 here replaces
ground motions. Similarly, the present study is performed using ten such motion time
to be used with Eq(3) varies
:quations (2), (4) - (7) yield histories, which are listed below
Parkfield, 1966, N5OE and N40W
(10) El Centro, 1940, NS and EW

close to the numerically El Centro, 1934, NS and EW

high frequencies. Taft, 1952, N21E and S69E

Olympia, 1949, S04E and S86W
For any pair of frequencies f1 and f2, the mean and the standard deviation of £12 were
calculated using the responses from the ten time histories.
When f2 is infinite, R2 (t) is equal to scaled input motion time history. For any
in the present study. The frequency fl, the factor al, is then given by C12 as stated in Eq(ll). The mean values of
action of the critical damping a were calculated for various frequencies and damping ratios, and were found to be in good
alues of a, which was found to agreement with Eq(12).
alues Fig 1 shows the comparison of c12 values calculated using Eq(10), and those obtained
(12) from the average of the values calculated using the ten motion time histories, for the
should be taken equal to 1.0, critical damping ratio of 1%. In Eq(10), actually calculated mean values of a's vere used.
ould be set equal to -0.:. As shown, the agreement between the calculated values and those given by Eq(10) is good.
ratio and are given by The agreements for other damping ratios (2, 4 and 7%) were also found to be good. This in
'1
effect verifies the assumptions made in this study, viz, a modal response have twc
components: rigid and damped periodic and that they are uncorrelated. When the frequency
f is small, it is seen in Fig 2 that for certain range of f2 values the correlation
(13)
coefficients have a small negative value. The writers feel that perhaps these negative
values can be attributed to a limited data in terms of using a small number of earthquake
two single degree of freedom
time histories. In any case, since these values are numerically quite small, writers did
bjected to the same earthquake
not attach much significance to them, and decided not to investigate them further.
covariance a12 are defined as
Next, the effect of any error in the value of c12 is investigated. It is noted that
the value of the combined response would be most sensitive to any errors in the vslues
of the correlation coefficents, when there are only two significant modal responses and
when they are also equal in magnitude. The value of the combined response is ther given by

R 2Rl x (1 + 12 ) R R2 (16)

.
If a different value of the correlation coefficient, c£l were used, a new value cf the
(14) combined response, R', would be obtained. The percentage change in the value of the
combined response is given by

ARZ-( +: -1) x100 (17)

(15)
There are two types of variations in c12 of interest here. One is the dispersion in the

:o the value of the duration value of £12 represented by the standard deviation. The other is the error introduced

ictically independent of the because of using the idealized equations such as Eqs(l0) and (12), which attempt to

; are covered in the assured calculate the mean value of c12 In order to investigate the accuracy of Eqs(10) and (12),
7
:snot pursued any further here. and the effect of dispersion in the value of c121 Eq(l ) is used to calculate the maximum
error in the response value by substituting in the denominator the value of c12 from

7/5 | -5- K 7/5

I1

--
: i. i~a;In
 1
Eqs (10) and (12), and in the numerator substituting for cli the calculated mean +bne
standard deviation. The maximum of all AR values for f2 /f1 - 1 to 10 for four fI - ty _ Eq(10)
!;'r-"
frequencies are given below. X Numerical

,:.r ,l>
Frequency 2 Error at
Maximum
1 .01 .02 .04 .07 1 = O. Io oHil

0.1 11.0 10.5 9.8 8.8

1.0 3.4 3.9 3.5 3.4
10.0 5.0 6.8 7.1 5.6
21.5 4.6 2.0 0.5 0.0

The highest error of 11% is encountered at the lowest frequency and damping. Generally, i U,

as the frequency and the damping go up the error diminishes until it becomes zero. Since
r
the worst cases are considered, in most practical situations the actual errors are likely z
to be much smaller.
P0
4. Conclusiois
1.
A modal response can be assumed to consist of two components which are uncorrelated,
Whereas the all the modal rigid
FS:0UENCY RRTIO F2/FI
a rigid component and a damped periodic component.
components are perfectly correlated, the correlation between the modal damped periodic
components is atfunction of the modal frequencies; an expression for the correlation
coefficient £1P, which is similar to one given in Ref[l], is given, Eq(8). On this basis
then, thiecorrelation coefficient for modal responses is evaluated which is given by Eq(l0).
F= . 10.00 H2
It is shown that the theoretically calculated modal correlation coefficients are in good
agreement with those calculated directly from the response time histories of single degree ;
of freedom systems subjected to actual ground motion histories. This in effect verifies
the assumptions made in the present study. 1,
5. Rfrne a-i
[1] Rosenblueti, E. and Elorduy, J., "Response of Linear Systems to Certain Transient r'i na
Disturbances," Proc., fourth World Conference on Earthquake Engineering, Santiago,
Chile, 1963, A-1, pp. 185-196.
oz
[2] Kennedy, R.. P., "Recommendations for Changes and Additions to Standard Review Plans andre
Regulatory Guides Dealing with Seismic Design Requirements for Structures," Report
prepared for Lawrence Livermore Laboratory, Published in Nureg/CR-1161, June, 1979.
(Also personal communications with Kennedy)

[3] Gungor, I, "A Study of Stochastic Models for Predicting Maximum Earthquake Structural FREQUENCY REITIo F?/F
Response," Ph.D. Thesis, University of Illinois at Urbana - Champaign, August, 1971.

[4] U. S. Nuclear Regulatory Commission, "Design Response Spectra for Seismic Design of
Nuclear Power Plants," Regulatory Guide 1.60, Revision 1, December 1973.
TI
Figure I - Comparison of
with Eq(l0), 5 - 00 O

- . I

I..,,..
I.
-6- K 715
the calculated mean + onl
1 to 10 for four f
Eq(10)
X Numerical

.07

8.8 Fl = 0. I000HZ Fl - 1.000 HZ

3.4
5.6 I
P
0.0
A
CD
y and damping. Generally,
.0,
til it becomes zero. Since
-0
r
(-I z
r t
he actual errors are likely ZCD zC 'D

nts which are uncorrelated, I1. 2 9 . 9. I10 2. q. 8. 15

e all the modal rigid
I FREQUENCY RATIO F2/FI FREQUENCY RATIO F2/F;
he modal damped periodic
on for the correlation
iven, Eq(8). On this basis
2ted which is given by Eq(10).
coefficients are in good
Fl - 10.00 HZ Fl - 21.5q HZ
3 histories of single degree
This in effect verifies

0 - UP
co

!ms to Certain Transient (I

ce Engineering, Santiago, -n3 To
C,, 0l)
r,
ZC ZC r
i to Standard Review Plans and
i for Structures," Report !D -

lureg/CR-1161, June, 1979.

P-I p
2. fI. 10 1. 2. i. S. 10.
tximum Earthquake Structural
- Champaign, August, 1971. FREQUENCY RATIO F2/FI FREQUENCY RATIO F2/Fl
tra for Seismic Design of
December 1973.

Figure 1 - Comparison of Numerically Calculated Correlation Coefficients, c

with Eq(10), C - 0.01

I :
I , I

K 7/5
l. . -7- K 7/5

II '
.1 -
s:&; '..1