Andjelkovic Vladimir Metal

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

Journal of Wind Engineering and Industrial Aerodynamics, 36 (1990) 829-843 829

Elsevier Science Publishers B.V., A m s t e r d a m - - Printed in T h e Netherlands

Stochastic Methods in Wind Engineering

Masanobu Shinozuka,lC-B. Yun,2and H. Seya 3

ABSTRACT

Two rather disparate issues, although both are stochastic in nature, arising
from wind engineering are considered in this paper. The first issue involves
methods for digitally generating sample functions of a wind velocity field. Such
digital generation of sample functions is a prerequisite to successful implementa-
tion of the time and space domain analysis of wind-induced structural response.
This is particularly true within the context of Monte Carlo soluations to various
problems not amenable to frequency and wave number domain analysis. The sec-
ond issue is concerned with the reliability analysis of building structures subjected
to multiple natural hazards (seismic and wind in this case) within the framework
of contemporary probabilistic risk assessment procedures.

INTRODUCTION

In the present paper, two rather disparate issues are considered, although
they are both stochastic in nature. The first issue involves methods for digitally
generating sample functions of a wind velocity field. The capabilities of such
digital generation is essential in performing time and space domain analysis
of wind-induced structural response within the context of finding Monte Carlo
solutions to various problems of wind engineering. These problems arise, for
example, from structural nonlinearity, parametric excitation and fatigue damage
sustained by structures under random loading. In many cases, the Monte Carlo
solution performed in the time and space domain is quite a competitive, if not
the only, analytical option available to the engineer who must find a solution on
the basis of which design decisions must be made, reliability analysis performed
and eventually risk assessment carried out. Specifically, the present paper is
concerned with a method for generating sample functions of a wind velocity

1 Professor, Dept. of Civil Engineering and Operations Research, Princeton University, Princeton, NJ 08544, USA.
2 Professor, Korean Advanced Institute for Science and Technology, Seoul, Korea.
3 Visiting Research Engineer, Princeton University; On leave from Takenaka Komuten Co., Tokyo, Japan.

0167-6105/90/$03.50 © 199if--Elsevier Science Publishers B.V.


830

field at a number (say, m) of specified points within the space. The method is
based on the assumption that the fluctuating components of the wind velocity
field can be idealized as a mean-zero multivariate and multidimensional Gaussian
process. For example, if the sample functions of all three components of wind
velocity at each spatial point are to be generated, a (3m)-variate Gaussian process
must be dealt with involving a (3m) x (3m) cross-spectral density function matrix
or cross-correlation function matrix. In the present paper, only one component
of the wind velocity is considered for digital generation, for simplicity. A
number of methods have been proposed for such sample function generation and
successfully applied to a variety of structural problems. These methods include
use of the auto-regressive and moving-average (ARMA) models, for example, by
Samaras et al. (1985) and Naganuma et al. (1987) and the spectral representation
method developed by Shinozuka and his associates as examplified by Shinozuka
(1986) and Yamazaki and Shinozuka (1988), the latter extending the method so
as to accommodate multidimensional non-Gaussian process cases. The spectral
representation method is usually used in conjunction with Cholesky decomposition
of the target spectral density matrix as part of the sample function generation
procedure. The alternative method presented here utilizes modal decomposition
of the spectral density matrix and appears to be conceptually more attractive,
particularly for cases with a large value of m, in which only a limited number of
modes may be utilized to achieve the necessary degree of accuracy in the sample
function generation. A numerical example is also worked out in the present paper
in which the longitudinal components of the wind velocity at three points on
a horizontal straight line perpendicular to the wind direction are generated as a
tri-variate Gaussian process.

The second issue deals with the reliability analysis of shear-wall building
structures subjected to multiple natural hazards within the framework of the
contemporary probabilistic risk assessment procedure. In the present study,
seismic and wind hazards are considered for the reliability analysis of conventional
structures, particularly low-rise shear-wall buildings. These structures are usually
designed according to provisions of building codes and standards which express
the wind and seismic forces as equivalent static forces, although their very nature is
random and dynamic. Studies of a similar nature have been carried out by Hwang
et al. (1988) for five-story shear-wall buildings. The present paper extends this
study to include shear-wall buildings with a maximum height up to 30 stories. The
results are shown primarily in terms of diagrams where the limit state probabilities
are associated with seismic and wind design.

G E N E R A T I O N OF MULTIVARIATE GAUSSIAN PROCESSES

Following primarily the procedure developed by Shinozuka (1987), consider a


set of m stationary Gaussian random process fo (t) (j=l,2 ..... m) with mean zero

E [i ° (t)] : o (1)
831
and with one-sided target cross-spactral density matrix S o (w) defined by

S~I (6d) S~2 (W) S0m (W) "]


"'q~l(W) S22(W) S2~m(W)[
s o @) = 0 < w < oc (2)
/
S0ml"(¢0) S°2"(~') S°t,n (g,O)J
where S°k (w) is the Wiener-Khintchine transformation of the cross-correlation
function R°k (r) (j # k) or the auto-correlation function R°k ( r ) ( j = k). The
matrix S O(w) is a non-negative definite and Hermitian matrix.

Suppose a matrix H (w) is found which possesses its Fourier transform and
satisfies the equation

S O(w) = H (w) I=IT (w) (3)

where the super bar indicates the complex conjugate and T matrix transposition.

Instead of the Cholesky decomposition used by Shinozuka (1987), modal


decomposition is carried out to obtain H(w) in the present study:

where ~(w) is the modal matrix of S°(w) and Aj is the jth eigen value. Modal
factorization is conceptually more attractive than the triangular factorization as-
sociated with Cholesky decomposition, particularly for cases with a large number
of variates (or large m). In these cases, only a limited number of modes may be
utilized in order to achieve the necessary degree of accuracy in the statistics of
the generated sample functions.

Once H(w) is computed using Eq. 4, the proceses f°(t) can be simulated
by means of the following series.

fj(t) = E [HJ k(Wl)N/2AW cos[w# + 0jk(wl) + '~kt] (5)


k=l I=l

where Aw = frequency inteval, wt = Aw(1 - 1), N = number of frequency


intervals, ffkt = independent random phase angles uniformly distributed between
0 and 2rr, and
[~{gik(~t)}
0jk(wt) = tan -1 [~{gjk(Wt)}] (6)

where ~{-} and ~{.} indicate the imaginary and real parts.
832

It then follows that the ensemble average E[fi(t)] is zero,

E[fj(t)] = 0 (7)

and the cross-correlation Rip(r) is given by


N
(s)
k=l l=l

By letting N -4 c,z, Eq. 8 reduces to Eq. 9.

Rip(r) = lira E Hjk(Wt)[-lPk(Wl)Awei~lr


N~oo

/?m
k=l l=l

k=l
= n;,,(,) (9)
Thus the process fj(t) (j = 1 , 2 , . . . , m ) , as simulated by gq. 5 produces the
target cross-correlation function R°p(r) and cross-spectral density S°p(W) with
respect to the ensemble average.

Replacing ffkt in Eq. 5 by the i th sequence of their realizations ¢~il), the i th


sample history --f}i)(t) is obtained. It can then be shown that the temporal average
d

of the i th sample time history is equal to the ensemble mean, which is zero,

< .fJi)(t) > = E[fj(t)] = 0 (10)


However, the temporal cross-correlation function and cross spectral density are
not equal to the ensemble values,

(0
< Rjk(~-) >¢ R~k(T) < S~k(w
/,) ) > ~ S°k(~) (11)

Hence, the process fj(t) given by Eq. 5 are stationary, but not ergodic.

A set of fluctuating components fj(t) (j=l,2,3) of wind velocity are generated


at three different points as shown in Fig. 1. Points 1, 2 and 3 are located at
x = :cl = 0 ft, x = x2 = 50 f t and x = x3 = 200 .['L respectively. The
following analytical form for the 3 × 3 cross-spectral density matrix for fj(t)
due to Davenport (1961, 1968) is used here.

s0j(~) = t,L~ . I~1 exp(-~l~l. I~,jl) (12)

where ~ij = [xi - xj[ (i,j = 1,2,3), and it is assumed that L = 4000 f L
k = 0 . 0 3 , c = L/(27ru33) with u33 = 40mph, a = 0.02 see~ft.
833

For the generation of fj(t) (j=1,2,3), Eq. 5 is used with Hjk(w) based on
modal decomposition of the target spectral matrix (Eq. 12). Figure 1 shows
one set of generated wind velocities at the three points. Figure 2 shows three
components of the cross-correlation matrix, R l l ( r ) , R22(r) and R12(r), obtained
from the sample time histories; Ri~j obtained from the first set of sample functions,
RiAvl° from the ensemble average over the first ten sets.The results are compared
with the• target correlation. function R °.
12"
It can be observed that the temporal
correlauon functions obtained from the single set of sample time histories do not
agree with the target values very well, particularly at large values of time lag
r. However the ensemble correlation functions appear to statistically converge
to the target values as the sample size becomes as large as 10; In fact,R°j and
RiAvl° are indistinguishable in the figure. The same trend also exists for R33(r),
R13(r) and R23(r).

R E L I A B I L I T Y ANALYSIS OF SHEAR WALL BUILDINGS


UNDER M U L T I P L E NATURAL HAZARDS

Shear Wall Buildings

Shear wall buildings located in New York City are considered. Figure 3 shows
the plan and section of a building. In the present study, the building is assumed
to be fixed at the bottom and the height of each story is taken as constant. The
design values of loads are as specified by the provisions of ANSI A58.1-1982
(1982). Also, the limit states are defined by the base shear Q for simplicity. The
wind velocity pressure qz specified in ANSI A58.1-1982 is

qz = O.O0256Kz(IV) 2 (psf) (13)


where V is the basic wind speed at a reference height of 33 f t for exposure C and
I the importance factor. From the map of basic wind speeds in ANSI A58.1-1982,
V = 80 mph in New York City for a return period of 50 years and I is taken as
1.05 (Category I at hurricane ocean line). The velocity pressure coefficient Kz
varies with height and is given by
Z 2
I(z = 2.58(-7-)z (14)
~g

where z is height from the ground. When the height is 15 fl or less, z is taken
as 15 ft. For exposure B, a = 4.5 and zg = 1200 ft.

The design wind pressure Pz is determined by the following formula:

Pz = qzah{Cp(w) - Cp(L)} (15)


834

where Cp(w) and Cp(L) are the wind pressure coefficients for windward and
leeward walls, respectively, and in the present study, Cp(w) = 0.8 and C'p(L) =
-0.5 are used. Also, Gh is the gust response factor at a height h given by

ah = 0.65 + 3.65 2'35(D°)½ (16)


h 1
(~)~
where Do = 0.01 for exposure B. Using Eqs.13-16, the lateral loads acting on
each floor is computed and the design base shear Qw due to wind are obtained
by summing all the lateral forces.

The design base shear QF~ acting on the building due to earthquake as specified
in ANSI A58.1-1982 is given by

QE = Z I K C S W (17)
where QE= total shear force at the base, Z = zone factor, I = importance factor,
K = building system factor, C = numerical coeffcient, S = soil factor and W
= total dead load.

Although New York City is located in seismic zone 2 according to the zonal
map of ANSI A58.1-1982, zone 1 is used in this study for which the value of Z
is 3/16. The importance factor I and building system factor K are both taken as
1.0. The value of C is determined by
1
c - 15Jr (is)
in which T is the fundamental period of the building in seconds and computed
by the following formula:

0.05hN
T- ~ (19)
where hN is the building height from the base and N is the number of floors. D
is the dimension of the building in the direction of applied seismic forces. Soil
factor S is assumed to be 1.0. The dead load W of the building is calculated
assuming constant mass density for each floor.

In the present case, the shear wall is designed according to ACI Code 318-83
(1983) which specifies the following:
1.4D + 1.7L
0.75(1.4D + 1.7L + 1.7W)
CRn >~ { 0.9D + 1.3W (20a - e)
0.75(1.4D + 1.TL + 1.87E)
0.9D + 1.43E

where D = dead load effect (not to be confused with dimension D of the building),
L = live load effect, W = load effect due to wind (not to be confused with the
835

W for dead weight), E = load effect due to earthquake, ¢ = strength reduction


factor and Rn = nominal capacity.

For wind load, the shear wall is designed incorporating Eqs. 20b and 20c. It
is assumed that the critical section to be designed is at the bottom of the shear
wall and the vertical loads are supported by the frame structure. Under these
assumptions for base shear, Eqs. 20b and 20c become

CVf f > 1.3Qw (21)


where VW = nominal shear capacity of shear walls designed under wind loading
only. The design formulas for earthquake load are given by Eqs. 20d and 20e.
Again under the assumption that the vertical loads are resisted by the frame
structure, Eqs. 20d and 20e become

CVE _> 1.43QE (22)

Actual Loads and Resistance

The probabilistic model P~ for wind pressure Pz is assumed to be

Pz• = O.O0256C;I(~Ch(V
, ., • , ) 2 (23)
where V* is the wind speed at reference height 33 ft. From the analysis of
observed data (1947-1977) collected at La Guardia Airport in New York City,
Simiu et al. (1979) estimated that the annual extreme wind speed Va follows
a Type I extreme-value distribution with expected value equal to 50.3 mph and
standard deviation equal to 7.23 mph (COV = 0.14). In this study, it is assumed
that the median value 17' is the same as the mean, i e., 50.3 mph and the
logarithmic standard deviation flv of V~ is f l v = 0.14. This notation for the
logarithmic standard deviation is used throughout. The statistics of C~, K~ and
G~, are described by Ellingwood et al. (1980). The median values of these factors,
C~, Kz* and G~, are taken to be the same as the design values. In addition,
the following values are also chosen flcp= 0.12, fll(z =0.16 and flGh =0.11 as
suggested by Ellingwood et al. (1980). Note that the wind pressure profile is
always the same as that specified by K**; its random variation depends on flop,
ilK,, t a b and flv. The probabilistic base shear Q~v due to the wind is a product
of the wind pressure and the exposed area of the building. (The dimensions of the
building are assumed to be deterministic.) Thus, the variation of the base shear
is the same as that of the wind pressure

flow = [fl~p + fl]iz + fl~'h + 4fl~] 1/2 (24)


From the data presented above, flQW is determined to be 0.36.

The probabilistic seismic base shear acting on the building Q~:, is determined
by the following expression as given in ATC 3-06 (1984):
836

1.2S'W*
* 4* (25)
QE - R,(T,)2/3"
In the above equation, A* is the annual extreme peak ground acceleration (PGA),
which is usually assumed to follow a Type II extreme-value distribution according
to Ellingwood (1980):

FA. = ~:~[_(o_)_~]
IL
(26)
For New York City, the parameters # and ct are taken as 0.0135 and 3.14,
respectively as suggested by Hwang et al. (1984). Thus, Eq. 26 gives a COV
of A* equal to 0.626 and A* = 0.0152. In this study, A* is assumed to follow a
log-normal distribution with the same median value A* = 0.0152 and/32 = 0.575
corresponding to COV = 0.625.

Figure 4 shows a plot of the seismic hazard curves: the curve given by
Eq. 26 is shown by a dashed curve and the curve corresponding to log-normal
distribution by a solid curve. The log-normal assumption gives an nonconservative
estimate of the seismic hazards for very high values of A*. However, it produces
a conservative estimate of seismic hazards in the range of A* where the structual
capacity is primarily located. Ellingwood et al. (1980) recommends that /~w
be 0.10 and the median of 1~;* be 1.05 times the design value, where W* is
the probabilistic weight of the structure. The factor 1.2/(T*) U3 is for linear
dynamic response amplification. Based on the data collected by Haviland (1976),
the median of period T* is taken to be 0.91 times the computed value, and/3T is
0.34. The median value/~ * for the probabilistic (nonlinear) response modification
factor R* is assumed to be 5.5 a n d / ~ is taken as 0.4. Finally, the median soil
factor S* is taken to be the same as the design value, which depends on the soil
type. as is assumed to be 0.3.

From Eq. 25, and from the assumed log-normality of all the variables
involved, the median of the seismic base shear ( ~ is
~ ~

1.2S*W*
-* A* (27)
O z - k,(~,)2/3
Furthermore, flOE can be determined as flOE = 0.80 from the following equation:
•9 2
= ~ + ( )/j~ +/~'d~/-' (2s)

The structual capacity QR is affected by variations in material strength, struc-


tual geometry and workmanship. For low-rise shear wall structures, Ellingwood
and Hwang (1985) estimate that the median ultimate shear capacity of shear walls
Q/z is about 1.70 times the nominal capacity Vn and /3R = 0.18. The struc-
tural capacity of shear walls designed for wind and earthquake is then expressed,
respectively as
Qw = 1.71/~w t}~ = 1.7V,~ (29)
837

Sensitivity Analysis of Reliability


The limit state probability is used as a measure of the structual integrity. For
shear failure under earthquake, the limit state probability P f,QE may be defined as

QR < 1) = <rr -I'~(ORIO'*E) I


PLQE = Pr(~-7- (30)
-- --L 2 2 1/2 j

since QR and Q~ are both assumed to be log-normally distributed. In Eq. 30,


~[.] is the standardized normal distribution function. Similarly, the limit state
probability PLQW under wind load is

PLQW= ¢[
-b~(
2
ORl2O;v1/:]
)o (31)
(~QR+/~QW)
again under the assumption that QR and Q~, are both log-normally distributed.
Inplicit in this evaluation of limit state probabilities is the assumption that the
two shear walls in Fig. 3 have identical strength and that one-half of QE or Qw
will act on each of them.

Figures 5a and 5b show the variation of limit state probabilities with floor
numbers for earthquake and wind design. Figure 5a shows that the limit state
probability of shear wall buildings designed for zone I seismic load (but not for
code-specified wind load) decreases slightly with increase in floor numbers when
subjected to seismic hazard. Figure 5a also shows that the limit state probability of
these buildings increases dramatically as building height increases when subjected
to wind hazard. The cross-over occurs at around a floor number equal to fourteen.
This implies that shear buildings designed for zone I seismic load (but not for code-
specified wind load) have more than sufficient strength to resist wind hazard if
the buildings are lower than about fourteen stories. Otherwise, the siesmic design
cannot provide sufficient resistance against wind hazard. This observation is valid,
however, only for the basic reliability model constructed in the preceding section
with specific values assumed for OR, O,w,/~R, fie and flw and the corresponding
log-normal assumptions. Similarly for the basic model, Fig. 5b shows that (a)
shear wall buildings designed for code-specified wind load (but not zone I seismic
load) have almost constant limit state probabilities over the entire range of building
height considered (1 - 30 stories) under wind hazard due primarily to similar wind
pressure distributions assumed for both design and actual wind conditions, and
(b) these buildings have larger limit state probabilities under seismic hazard than
those under wind hazard, the difference being particularly large for buildings a
few stories high. These results for the basic model indicate that low- to medium-
rise shear-wall buildings (1 - 30 stories high) designed under the current design
code without siesmic provisions are highly vulnerable under seismic hazard.

CONCLUSIONS

The first half of this paper developed a method of digitally generating sample
functions of multivariate stochastic processes. The method involves the modal de-
838

composition of the cross-spectral density function matrix, rather than the Cholesky
decomposition often used. Sample functions thus generated are not ergodic (re-
gardless of modal or Cholesky decomposition), but an ensemble average of ten
sample functions appears to provide good agreement with the target auto- and
cross-correlation functions, at least for the type of wind velocity field considered
in this study.

In the second half, a reliability analysis is performed for shear building


structures in New York City. It is concluded that low-rise shear wall buildings
designed according to the code appear to have much higher limit state probabilities
under seismic hazard than under wind hazard. In this regard, it is strongly
recommended that a sensitivity analysis of the estimated reliability values to
various uncertain factors involved in the analysis be carried out.

ACKNOWLEDGEMENT

This work was supported by the National Center for Earthquake Engineering
Research under Grant No. NCEER88-1003.

REFERENCES

American Concrete Institute, "Building Code Requirements for Reinforced Con-


crete," ACI 318-83, Detroit, Michigan, 1983.

American National Standard Institute, "Miimum Design Loads for Buildings and
Other Structures," ANSI A58.1-1982, New York, 1982.

Applied Technology Council, "Tentative Provisions for the Development of Seis-


mic Regulations for Buildings," ATC 3-06, Redwood City, California, 1984.

Davenport, A.G., "The Spectrum of Horizontal Gustiness Near the Ground in


High Winds," J._z-Royal Meteorol. Soc., Vol. 87, pp. 194-211, 1961.

Davenport, A.G., "The Dependence of Wind Load Upon Meteorological Param-


eters," Proceedings of the International Research Seminar on Wind Effects
on Buildings and Structures, University of Toronto Press, pp. 19-82, 1968.

Ellingwood, B., Galambos, T.V., MacGregor, J.G. and Cornell, C.A., "Develop-
ment of Probability-Based Load Criterion for American National Standard
A58," Special Publication NBS SP 577, National Bureau of Standards,
Washington, DC, 1980.

Ellingwood, B. and Hwang, H., "Probabilistic Descriptions of Resistance of


Safety-Related Structures in Nuclear Plants," Nuclear Engineering and
Design , Vol. 88, pp. 169-178, 1985.
839

Haviland, R., "A Study of the Uncertainties in the Fundamental Transitional


Periods and Damping Values for Real Buildings," MIT, Cambridge, Mass.,
1976.

Hwang, H., Ushiba, H. and Shinozuka, M., "Reliability Analysis of Coded-


Designed Structures Under Natural Hazards," Technical Report NCEER-
88-008, 1988.

Hwang, H., Shinozuka, M., Kawakami, J. and Reich, M., "Reliability Assessment
of Indian Point Unit 3 Containment Structure Under Combined Loads,"
Structural Engineering In Nuclear Facilities, ASCE, Vol. 1, pp. 274-293,
1984.

Naganuma, T., Deodatis, G. and Shinozuka, M., "An ARMA Model for Two-
Dimensional Processes," J_ Engineering Mechanics, ASCE, Vol. 113, No.
2, pp. 234-251, 1987.

Samaras, E., Shinozuka, M. and Tsurui, A., "ARMA Representation of Random


Processes," J~ Engineering Mechanics, ASCE, Vol. 111, No. 3, pp.
449-461, 1985.

Shinozuka, M., "Stochastic Fields and Their Digital Simulation," Stochastic


Methods in Structural Dynamics, Eds. G.I. Schueller and M. Shinozuka,
Martinus Nijhoff Publishers, Boston, pp. 93-133.1987; also in Stochastic
Mechanics, Vol. I, Ed. M Shinozuka, Columbia University, 1987.

Simiu, E., Changery, M.J. and Changery, J.J., "Extreme Wind Speeds at 129
Stations in the Continental United States," Building Science Series NBS
BS 118, National Bureau of Standards, Washington, DC, 1979.

Yamazaki, F. and Shinozuka, M., "Digital Generation of Non-Gaussian Stochastic


Fields," J. Engineering Mechanics, ASCE, Vol. 114, No. 7, pp. 1183-1197,
1988.
840

5O

-g

,-,...,
, p
-~0 i~ 3~ 4~Q
x
4
-a (e, ~" so

T I

5O

"D

J- a~ 4~0
Time (sec.)

Fig. 1 Generated Sample Functions of Fluctuating Components of" Longitudinal


Wind Velocity
841

200

....a~ I

¢,,i

-200 T T r T
50 1 oo 150 200 250

200

__.~0
I'-

-200 I 1 ! I I
0 50 100 150 200 250

200

53
-200 I I I
0 50 100 150 200 250
"r(sec.)

Fig. 2 Three Components of the Cro6s-Correlation Matrix; R~I(r), R22(r) and R12(r)
( R°1 = Target, R~I = Sample 1, RA~ 1° = Average of 10 Samples )
842

I. b
-I
Shear Wall

I I t ~.+,i:0 D

-y

nm

b 125 ft. h 13 ft.


D 75 ft. PD 0.12 kips/re=
Figure 3 Plan and Section of Building

!
i0 ° ~ j , i , , i r , + + , I I ' + ' +' '

~J lO-i
g
X

,.~ IQ "z

: :1 " ' - # = 0.0135


I I0"

.<
r 0151719 ; t t---
IO-I"
0 .oz .o~ .05 .oa .~o .z~ .~ .J6 .,s .20 .~

Peak a c c e l e r a t i o n a ( i n g)

F i g u r e 4 Seismic H a z a r d C u r v e
843

I0 o
l

Jo 10 -5

lO-tO

lO-~S
0 5 I0 15 20 25 30

Number of floors

Figure 5-a Limit State Probability Design for E a r t h q u a k e

I0 o
I I

10 -5
J Wind

. 10-1o
E

[ ' I t i i ~ t ' ' t P ' ' ' T


I0-~5
0 5 I0 15 20 25 30

Number of floors

Figure 5-b Limit S t a t e Probability Design for Wind

You might also like