Professional Documents
Culture Documents
1 s2.0 0167610596000037 Main PDF
1 s2.0 0167610596000037 Main PDF
winde~neednt
Journal of Wind Engineering &(~]~i~
ELSEVIER and Industrial Aerodynamics 59 (1996) 115-130
Abstract
This paper proposes an RDT ranked by peak amplitude for directly and effectively evaluat-
ing the amplitude dependences of dynamic phenomena. It first theoretically explains the
principle of RDT based on a conditional expectation. Then, it shows that a modified condi-
tional expectation enables the RDT to estimate amplitude dependences, and derives the RDT
ranked by the peak amplitude. Analytical and numerical examples are examined for a simple
SDOF model. The proposed technique is then applied to wind-induced response data of three
towers and the results are compared with those of the traditional technique.
1. Introduction
*Corresponding author.
1Professor.
ZResearch Associate.
2. Principle of RDT
where E {xlC} is the conditional expectation of.v tinder the condition C and sgn [x] is
the sign of.v. 2{t) denotes the derivative ofx(t} with respect to time t, and thus 2(¢) = 0
means that .v(/) has a peak value, say .¢:. at time t. Suppose the response x(t) is
a stationary Gaussian process with mean 0. Then
where .v - N(it, a-') means that the random wtriable .x" is Gaussian with mean p and
variance a-'. R,.,.{r) is the auto-correhition function of x(¢). R,..,tr) and Rx.,.{v) denote
the first and second derivative of R,.,(Tt with respect to the time lag T, respectively.
Therefore, we can write the R a n d o m d e c signature a{r) defined by Eq. (1) as
R,.d'c)
a(r) = - - " El sgn [2] 21 . (3)
R,.,(0!
alr~ -- E I s g n [ 2 ] ' 2 1 e (
......- c o s \ I --~'-'u;.r +
\ l ~_~2 sinx, 1 - ~ - r % r ) . (4)
Y. Tamura, X Suganuma/J. Wind Eng. Ind. Aerodyn. 59 (1996) 115-130 117
Therefore, we can estimate the natural frequency C0oand the damping ratio ~ by fitting
the Randomdec signature a(r) in the form of Eq. (4) in the least-squares sense. The
popular and convenient alternative to evaluating the damping ratio ~ is to solve
where fi~ is the ith peak value of the Randomdec signature a(~) for ~ >~ 0.
As it is impossible to evaluate the conditional expectation using only a sample of
response x(t), Eq. (1) is of no practical use in itself. Assuming that the response x(t) is
ergodic, we replace the expectation with the time average
1 N
a(~) = ~ sgn[x(ti)] x(ti + ¢) , (6)
i=l .4¢(t i ) = 0
where the summation on the right-hand side denotes the conditional summation with
respect to t~ which satisfies :~(t~) = 0.
The following simple model is considered to show the significance of Eq. (7). The
model represents the time-varying SDOF system in which the natural frequency
~Oo and the damping ratio ~ depend on the amplitude of the previous peak Xp. Thus,
the dynamic characteristics are constant between two neighboring peaks, i.e., the
118 Y. Tamura. ,5. Su,gammm../. Uind EnV. lnd .4ero~tvn. 59 (1996) 115 130
= ~,, + ¢ l X p . (9)
Strict calculation of Eq. (7) is very difficult, so the interval between two neighboring
peaks is assumed Io be half lhe current natural period. Therefore, the amplitude
dependence of the natural frequency ires no effect on the discussion. Using the
characteristics of the Gaussian randorn variable and the first-order approximation of
the exponential function, we obtain
where di(2) is the ith peak amplitude of the R a n d o m d e c signature a(2: r) for r ~> 0.
Substituting Eq. (10) into Eq. (5) for several peak values 2, the analytically estimated
damping ratios are calculated as shown in Fig. 1. In this figure, each marker denotes
the relation between the ith peak wflue of the R a n d o m d e c signature a(.',; r) and the
d a m p i n g ratio estimated using the ith half period. The estimated d a m p i n g ratios tend
to be overestimated as i increases and this trend is remarkable at small peak values.
Note that the R a n d o m d e c signature a(2; T) is neither a free-vibration response of the
time-varying model nor one of the time-invariant system with damping ratio ~(.~).
...................................
i.. I I I I I I A A A A A A A ..-
i I/VVVVVVV,,, ,.
~- 5 ..........~ .........................i.~ 7 ...... ~. . . . . . : ..-
* I ....... ~ = l+Xp[%l ]
O
c~ i ,\', * 1.00
........~].~alD.......... . . . . (~.25 • 1.2.s ,-
"E
r-,
3
, i i i J ~ J L L [ J l ~
c).5 ] !5 2
,\mpliludc
As the damping ratios estimated using the first half period coincide with the definition,
it is possible to evaluate the amplitude dependence by using the beginning of the
Randomdec signature a(2; r). We note that the estimated damping ratios must be
overestimated in any case.
As Eq. (1), Eq. (7) is of no practical use in itself. Assuming that the response x(t)
is ergodic, we replace the expectation with the time average. It is almost impossible,
however, to pick up many samples which satisfy the condition in Eq. (7), i.e., peak
values equal to 2. Consider the moderate condition which requires the magnitude of
peak value 2 to be within a particular range, rank R.
Then the RDT ranked by the peak amplitude is derived as
The amplitude dependences of the dynamic characteristics can be evaluated using the
Randomdec signatures calculated by Eq. (11) for many ranges R. In the following, lhis
procedure is called procedure A.
where X [10- 2 m] denotes the mean amplitude of the previous and subsequent peaks.
The above equations refer to the measurements by Tamura et al. (1993). The simula-
tion consists of the following steps:
(i) simulate a stationary random process which represents the fluctuating wind
force,
(ii) calculate the along-wind displacement of the model x(t),
(iii) remove the components not concerned with the fundamental mode using
a band-pass filter,
(iv) pick up all peak values and divide them into sever~il ranks based on order of
magnitude so that each rank consists of 5000 peak values,
(v) calculate the Randomdec signature a(R; ~) for each rank R using Eq. (11), and
(vi) estimate the natural periods and the damping ratios using the least-squares
approximation.
The results are shown in Figs. 2a and 2b with respect to the natural period and the
damping ratio, respectively. In this figure, each marker denotes the relation between
120 ). T a m u r a , S. S t g a n u m a ',1, l l T n d E n g . I n d . . 4 e r o e t v n . 59 (1996/ 115-130
1.25 F- . . . . . -
: ! :~ 5 /.2r,,
1.20 ........ ;; .......................... i
[]
iA AAA
Uvvvvvv
~i+4
v'
[] i - Randomdec signature
1.15
i i=1
................................:..................................i . , . . , m . . . , . . m . . . . . . , . ~ ............................. :
g:a.
Z 1.05
1.00
, - ...... h, - i + (1._08 X sec
Estimated using the first two periods [
0.95
0 0. I () 2 11.3 0.4 ()y
Amplitude l lO 2 nil
(a)
10 ] - - -
, , , , I , ~ t , I , i i i i i i L i i i L
the ith peak value of the Randomdec signature a(R; r) and the dynamic characteristics
estimated using the two periods beginning at the ith peak. Because of the time average
in Eq. (11), the variation of the Randomdec signature a(R; T) requires more than a half
period to estimate the dynamic characteristics. Note that the estimated damping
ratios shown in Fig. 2b have the same tendency as the analytical results shown in Fig.
1. The solid lines in Fig. 2 represent the dynamic characteristics estimated using the
first two periods of the Randomdec signature a(R; r), which reproduce the amplitude
Y. Tamura, S. Suganuma/~ Wind Eng. Ind. Aerodyn. 59 (1996) 115-130 121
dependences defined by Eqs. (12) and (13). It is recognized that the damping ratios are
overestimated, as expected, and that the natural periods are also overestimated.
Rxx(Tm) = 0. (15)
Then £ means approximately the mean amplitude of the half period beginning at time
t. Based on the same assumption as for Eq. (2), we obtain
Rxx(z) - Rxx{~ - Tin) _
a(£; r) = x. (16)
Rxx(0) -- Rxx(Tm)
In particular,
a(.~; Tm) = - - 2~. (17)
It follows that the first half period of the Randomdec signature a(2; r) shows no
damping at all. Similarly, it is easy to see that the extended Randomdec signature
which includes the condition about x(t+ T) differs from a free-vibration response
before time T. Therefore, the RDT ranked by the mean amplitude makes no sense.
As stated above, we can estimate the natural frequency coo and the damping ratio
by fitting the Randomdec signature a(z) in the form of Eq. (4) in the least-squares
sense. Assuming that
E {sgn[2] ~} = a(0) (18)
in Eq. (4), the minimizing variables become only COo and (. However, the auto-
correlation function just approximates a free-vibration response for z > 0 and the
Randomdec signature a(z) obtained by using the time average changes. In addition,
fi(0) does not always equal 0 because of the sampling for practical calculation.
Therefore, the minimization with reduced variables sometimes leads to a curious
result, e.g., the evaluated amplitude dependence has an imaginary jump, as shown in
Fig. 3. Using the generalized form with four variables,
where A and B denote dummy variables, we can obtain precise estimates at the cost of
processing time.
122 }. Tamura, S, Sugam#na,,]. ll)'nd E/lg Ind. Aerodyn. 59 (1996) 115 130
0.77
iI ! ~e Estimatedusing2 parameters~
i.~ -~ Estimatedusing 4 p a r a n ~
= 0.76
g
¢3"
: ~ ~ i * •~•e
"• "v'hD'G~ • i
0.75
........ ~ _ ~ : ; ~ 5 ~ % 7 , ............ " ........... g . . . . . . . . . . . . . . . . . . . i ....
Z
0.74
0.2 0.4 0.6 0.8
Arnplitude 110-2 m/s21
I¢1 E
R2[
a (R2;r),~r
(a)
a(r) a(r)
Vvvv vv 'vvVUl
RMS : ~'~
t vv"vV
R M S : a2
Vv,,vvv~.,
4.1. Tower A
I s.15 m A
Y
~"-~x - ~ N
(b)
] [5.15 m
24 m [ (a)
1.05
:2
• • 0o • • •
Z
! •
!
1.03 . . . . . i , , , ~ . . . . . . i , , ,
0 f).2 0.4 06 0,g I ._
,Alllpllttlde [ ]0 2 m/,,2
(ai
2,0
L
• Procedure A
,~ Procedurc B
&" 1.6
1.2
'K.
............. .~:
_ ~ ~: t . . , ., .........., ............., . ............................... ° ........
• ~e ••
.o,/~ ~,o o 0=
0.4
0.2 0.4 0.~ 0.8 .2
,,\mplimdc [1(}-2 m/~ j
(t,)
Fig. 6. Estimated dynamic characteristics (lox~cr A \ dircction): (a) natural frequency; Ib) damping ratio.
1.05
ooo!
O~ go
i
e. D o ° i
i oO~ • i
1.04
1.03
0
, , ,
0.2
I ~ , , I
0.4
, , I .
,
0.6
i . . .
0.8
Amplitude [10-2 m/s2]
. .
2.0
o [ • Procedure A ]
o o o Procedure B
1.6
o
o oo
~o 1.2 ............
o oO
6 ............i............................
z ..................................................................................................................
"~, oo i i o
E o :
o! i
0.4
0 0.2 0.4 0.6 0.8
(b) Amplitude [ 10 2 m/s2 ]
Fig. 7. Estimated dynamic characteristics (tower A y directiont: (a) natural frequency; (b) damping ratio.
procedure B, and each circle is evaluated using 2000 peaks. The natural frequencies
tend to decrease as the amplitude increases in both directions, although the natural
frequency in the y direction is about 0.005 Hz larger than that in the x direction. The
damping ratios tend to increase with the amplitude, with a few differences between the
two directions. The amplitude dependence of the damping ratio estimated by proce-
dure A nearly coincides with that evaluated by procedure B only in the x direction.
The latter is not clear in the y direction because of the scattered distribution. It is
recognized that procedure A produces better resolution than procedure B.
4.2. Tower B
(a)
[]
~.'3m I 22.0m I
(b) (west view) south view)
0.77
0,76
0.74
0 0,2 04 0.6 0.8
( ) Amplituclc> 11()2 m/s? I
1.6 i o
1.4
.~ 1.2 - .................
_ ~ * . . 5 . ' ; .................. .........................................................
Fig. 9. Estimated dynamic characteristics Itower B \ directionl: (al natural frequency; (b) damping ratio.
Figs. 9 a n d 10 s h o w t h e e s t i m a t e d d y n a m i c c h a r a c t e r i s t i c s in t h e x a n d y d i r e c t i o n s .
r e s p e c t i v e l y . T h e d i m e n s i o n s c a u s e t h e w i n d - i n d u c e d r e s p o n s e in t h e x d i r e c t i o n
t o b e a l w a y s d o m i n a n t a n d t h e r e is ilo s i g n i f i c a n c e in c o m p a r i n g t h e r e s u l t s for t h e
t w o d i r e c t i o n s . T h e circles r e p r e s e n t t h e restllts for p r o c e d u r e B a n d e a c h circle is
l( Tamura, S. S u g a n u m a / J . Wind Eng. Ind. A e r o d v n . 59 (1996) 1 1 5 - 1 3 0 127
0.98
evaluated from 1 hour of data, corresponding to about 3500 peaks. Note that the
results using procedure A are obtained from much fewer data than those using
procedure B. The natural frequencies tend to decrease as the amplitude increases in
both directions and, at the same time, the declination diminishes as the amplitude
increases. The damping ratio in the x direction increases with the amplitude and then
becomes almost constant for large amplitudes, while the damping ratio in the
y direction is nearly constant regardless of the amplitude. The amplitude dependence
of the damping ratio estimated by procedure A nearly coincides with that evaluated
by procedure B. Again, it is recognized that procedure A produces better resolution
than procedure B.
4.3. Tower C
Tower C is a steel-framed observatory tower 99.35 m high, as shown in Fig. 11. It has
a 16.5 m square cross section in its lower third, a 12.1 m square cross section in its
middle third and an octagonal section in its upper third. The reinforced-concrete
foundation is supported by caisson-type piles. The exterior walls are made of glassfiber-
reinforced concrete panels. Responses were measured on the top floor of the tower.
128 g. T a m u r a , S. S u g a n u m a / J . k H n d Eng. Ind. A e r o d v n . 5 9 ( 1 9 9 6 ) 1 1 5 - 1 3 0
@
Y t 12,100 I
Y N
~.~ 16,500 I
(a) x I (h)
O. 68
• Procedure A
•..o.d~%..¢,.o.................................................................................. o Procedure B
0.66 • +
g i •
o :: •
0.64
i i i ~o
"~ 0 . 6 2
Z
0 4 8 l0 12 14 16
A m p l i t u d e I 1(t-2 m/s 21
(a)
4.0
o 3.0
c i :+ .................................................
2.5 ................... ~.....................
c~ o
~.........IL......
•
z...,.g
i
r .................................................................................................
i } i i
E 2.0 ................... ~......• ............ i ............... o...~ ..................... i .................. + ...... ~, .. i
: ooo ~ ' o +
1.5 +
I.~.....Q.....~% ................................................. ....................... ++ [ •
o
Procedure A
Procedure B
]
1.0 , , , I L + , , , , , , , I . . . . . . I , , , ] , , ,
0 2 6 8 10 t2 14 16
(I,) A m p l i t u d e I l0 2 m / s 2 I
Figs. 12 and 13 show the estimated dynamic characteristics in the x and y direc-
tions, respectively. The circles represent the results for procedure B and each circle is
evaluated from 10 min of data, corresponding to about 400 peaks. The natural
frequencies tend to decrease as the amplitude increases in both directions, although
Y. T a m u r a , S. S u g a n u m a / J . W i n d E n g . Ind. A e r o d y n . 59 (1996) 115-130 129
0.68
io • Procedure A
o Procedure B
~0.66 .......................................... ..., ......................... ..
0.60
0 2 6 8 10 12 14 16
(a) Amplitude [ 10-2 m/s2 ]
4.0
i o
)0 0 - - u : q i ~ ! -- - " - - -
1.5 i i ] • Procedure A 1
......................................... !'"'"o!......................................................... ii'[ o Procedure g 1'
1.0 , , , , , , I , , , , , , , , , l , i , , , , ] , , ,
0 4 8 10 12 14 16
Amplitude [10 .2 m/s2 ]
(b)
Fig. 13. Estimated dynamic characteristics (tower C y direction): (a) natural frequency; (b) damping ratio.
the natural frequency in the y direction is about 0.005 Hz smaller than that
in the x direction. The damping ratios tend to increase with the amplitude,
with a few differences between the two directions. The amplitude dependence of the
dynamic characteristics estimated by procedure A coincides in general with that
evaluated by procedure B in both directions. It is recognized that there is a qualitative
resemblance between the results for towers A and C. Note that the estimated
damping ratio shows a slightly different tendency between the two directions for
small amplitudes for both tower A and tower C, Since the dimensions, the condition
of the foundation, and the structural and architectural finishes are almost the
same in the x and y directions, the difference may be attributed to aerodynamic
damping.
5. Concluding remarks
The RDT ranked by peak amplitude was introduced for directly and effectively
evaluating the amplitude dependences of dynamic characteristics. The principle of
RDT was explained theoretically based on a conditional expectation, then it was
130 E Tamura. S. 5;uganuma,'.L Wind Eng. hul. .4erodvn. 59 (1996) 115 130
Acknowledgements
The response records were provided by Mr. R. Kohsaka, Azusa Sekkei, Dr. K. Fujii
a n d Mr. K. Miyashita, W i n d Engineering Institute Co., Ltd., a n d Mr. K. Shimada,
Shimizu C o r p o r a t i o n . The efforts of Mr. A. Sasaki a n d Mr. Y. Shoji are acknowledged
with thanks.
References
,leary, A.P. (1986) Damping ill tall buildings a mechanisnland a predictor, Earlhquake Eng. Struct. Dyn.
14, 733 75(/,
Jeary, A.P. 119921 Eslablishing non-linear damping chalaclerislics of structures from non-slationary
response time-histories, Struc(. Eng. 70, 61 66.
Tamura, Y., K. Shimada and K. Hibi (1993) Wind response of a lower Ityphoon obserwltion at Nagasaki
Huis Ten Bosch Domtorenl. J. Wind Eng. lnd Aerodyn. 50. 309 318.
Tamura, Y.. M. Yamada and H. Yokota (1994al Estimation of structural damping of buildings, m: ASCE
StructuralCongf. and lASS Int. Symp.,Atlanta. USA, Vol. 2, 1012 1017.
Tamura, Y.. R. Kohsaka, O. Nakamura, K. Miyashita and V.J. Modi (1994b) Wind-inducedresponses of an
airport tower efficiencyof tuned liquid damper, in: Preprinls. East European Conf. on Wind Engineer-
ing, Warsaw, Poland. Part 1, Vol. 3, 175 184.
Vandiver, J.K., A.B. Dunwoody, R.B. Campbell and M.I. ('ook 11982) A mathematical basis for the
random decrement vibration signature analysis technique. J. Mech. Des. 104, 307 313.