Professional Documents
Culture Documents
Galambos 1986
Galambos 1986
ABSTRACT: The reliability of floor and roof deflections due to occupancy and
snow loads, respectively, and of lateral frame deflections due to wind loads is
examined. These deflections are determined for unfactored code-specified loads,
Downloaded from ascelibrary.org by University Of Sydney on 09/20/13. Copyright ASCE. For personal use only; all rights reserved.
and they are not to exceed professionally established deflection limits. First-
order, second-moment probabilistic theory is used in the analysis. Values of
the reliability index are calculated for eight- and one-yr reference periods. Rec-
ommendations are made with regard to load levels to be used in the service-
ability analysis of deflection-sensitive and deflection-insensitive structures.
INTRODUCTION
67
limits depend on a host of factors that are difficult to define in the design
process.
Floor Deflection Limits.—In customary steel design practice in the
United States, a deflection limit under live load or snow load of 1/360
of the span is used when the steel beam supports a plaster ceiling. A
larger deflection limit, usually 1/240 of the span, is used for other con-
ditions (6). Some codes require a limit of 1/360 of the span for floor
deflections, while permitting a deflection of 1/240 of the span for roofs.
Other countries use similar limits, with minimum allowable deflections
seldom less than 1/360 of the span nor seldom larger than 1/180 of the
span (6). A comparison of these limits with Table 1 suggests that the
customary floor deflection limits used by the designers are reasonable
but imprecise indices of damage or unsightliness. Until a major research
effort is mounted on a worldwide basis to define better limiting criteria,
there seems to be little basis for changing the existing deflection limits
when adopting new design methods based on LSD principles. Thus the
customary time-tested deflection limits will be used herein as determin-
istic allowable floor deflection criteria. These limits are either specified
in building codes or are imposed by the designer based on judgment
for the particular type of building under consideration.
Drift Limits.—Lateral deflection or rotation criteria are imposed on
the design of multi-story buildings to control the displacement under
wind loads. Chapter SB5 of Ref. 2 gives a summary of current practice.
The normally accepted range of drift limits for building frames is from
a low of 0.0014 to a high of 0.004 times the building height. The range
of inter-story drift limits is from about 0.0015 to 0.005 times the story
height. Drift limits vary depending on the occupancy, the cladding type,
the framing, and the judgment of the designer. It is significant to note
that the drift limits are of about the same magnitude as the accuracy
with which the building can be erected. The successful performance of
most buildings seems to indicate that the usual drift limits do provide
69
arate investigation for motion is thus warranted for tall buildings. Until
future research indicates otherwise, the most reasonable course of ac-
tion, at the present time, is to accept the customary drift criteria as de-
terministic limits in a reliability analysis.
Serviceability Loads.—Allowable deflection limits are matched against
computed displacements in the design process. Deflection computations
are made for an idealized structure, the response of which to the spec-
ified "service loads" is determined by performing a first-order elastic
analysis. Mitigating features, such as beam end-restraints and cladding;
and detrimental effects, such as second-order (or P-delta) deflections,
are usually neglected in this analysis. The "service loads" generally are
understood to be the code-specified (unfactored) live, wind or snow loads
(1,2).
Probabilistic load models are necessary to evaluate serviceability cri-
teria from a reliability viewpoint. It has been customary in most reli-
ability analyses to seek a random variable representation of the struc-
tural load of interest, in lieu of the more complex stochastic process
representations (4). The maximum load to occur within a period of ref-
erence, T, is defined as Q,„ = max Q(t), in which Q(t) is the load which
varies as a function of time. Methods exist for approximating the statis-
tics of Q,„ (3,4,8). The period of reference T must be selected carefully.
Loading criteria for ultimate limit states in the United States (1) are based
on a 50-yr period of reference. It does not seem reasonable to base ser-
viceability criteria on such a severe requirement. Instead, load criteria
for checking deflections might be founded on the premise that the de-
flection limit should not be exceeded more than once, on the average,
during one tenancy. The average period between tenant changes in of-
fice buildings is eight yrs (1). If the serviceability limit state is recover-
able, i.e., there is no permanent damage following removal of the load,
the checking load might instead by chosen to ensure that the deflection
limit is not exceeded annually.
The statistical properties of the loads corresponding to one- and eight-
yr periods of reference were given in Ref. 3, and are reproduced in Table
2 of this paper. Mean values, as fractions of the code-specified nominal
values, and Coefficients of Variation (COV), are given for occupancy,
snow and wind loads. The nominal loads are given in a load code (1)
for a 50-yr reference period; the mean load ratios in Table 1 thus rep-
resent the proportion of the life-time maximum load expected during the
shorter serviceability reference periods. These data will be used in the
subsequent reliability analyses.
FLOOR DEFLECTION
The maximum center deflection of a uniform beam with stiffness EI
and length L under a uniformly distributed load w, having equal rota-
70
It can be shown (see Appendix I) that for practical beams the effect
of restraint (i.e., the value of the bracket in Eq. 1) is to reduce the beam
deflection to between 80 and 99% of the value for the simply supported
beam. The effect of this restraint on the reliability will be discussed later
in this paper.
The reliability analysis is based on the assumption that: (1) The allow-
able deflection 8„ is a deterministic value specified by a code or by the
designer; and (2) the variability of the flexibility F is negligible when
compared to the variability of the loads. This latter assumption is veri-
W
k^LLLLLLlJ:
rfm~
L
"(l)L- (5)
where L/L„ is the ratio of the mean to the nominal load L„. Such ratios
are given in Table 2 for occupancy and snow loads for eight-yr and one-
yr periods of reference (3).. Also given are the corresponding COV's. The
mean loads are obtained by considering current design practice, i.e.
FL„ £ 8„ (6)
Substituting L„ = ba/F into Eq. 5 gives, with Eqs. 4 and 3, the following
expression for (3:
1
-1
L
L„ -
p= (7)
VL
It it is desired to include the end-restraint, then
p = ^ (8)
VL
where Y is the bracket in Eq. 1, i.e.
0.4
Y = l
~m <9>
— + 0.5
kL
Note that the reliability index depends only on the load statistics, L/Lm
and VL, and not on the deflection limits. This is a consequence of as-
suming that the beam deflects precisely to 8„ under L„ (Eq. 6).
The results of this analysis are given in Table 3 for restraint-effects Y
varying from 0.8-1. This range covers the restraint to be expected in
72
nominally simply framed beams. The reliability indices listed are for an
analysis which neglects the variability of the flexibility (these are the val-
ues in parentheses). The other values of p are determined in Appendix
I where the flexibility is taken to be a random variable. As can be ex-
pected, the latter (3's are somewhat smaller. However, the general trends
are essentially the same whether or not the flexibility is considered to
be random or deterministic.
Most simple beams probably fall into the range of Y = 0.9 to 0.95, and
it can be seen that for a floor designed for a customary live load deflec-
tion of, say, 1/360 of the span, the reliability index is somewhere be-
tween 1.6 and 2, considered for a reference period of eight yrs.
Table 3 also shows that the reliability index for current deflection ser-
viceability designs is larger for a period of reference of one rather than
eight yrs, and that roofs under snow load appear to have more reliable
performance than do floors under live load.
This disparity suggests a possible improvement in current practice, as
follows: If it is assumed that current deflection limits are satisfactory for
inhibiting damage and visual annoyance (this does not include using
deflection limits to control vibration or corrosion—see previous discus-
sion), and if it is postulated that serviceable performance is tantamount
to limiting the probability of exceeding the deflection limit for the period
between occupancy changes, i.e., eight yrs, then approximately the same
level of reliability could be achieved for the live and snow load deflection
checks by taking the loads at the following percentages of their nominal
values (1):
LATERAL DEFLECTION
Lateral drifts are determined in current design practice for the unfac-
tored nominal wind forces (1). The drift is computed either by first-order
or by second-order elastic analysis. The former method uses the actual
lateral loads, while the latter includes the additional lateral forces due
73
to the P-delta effect, i.e., the total lateral force is the sum of the wind
force and the force p n 2P, where p„ is the allowable drift limit and EP is
the total gravity load assigned to the story (see Fig. 2).
Appendix II gives the derivations for the determination of the reli-
ability index p for single-story frames. The resulting values of (3 in Table
4 indicate that for all practical purposes the inclusion of the variabilities
due to the flexibility and the gravity loads do not affect the reliability
significantly. Thus the following discussion will consider the lateral wind
load to be the only random parameter.
74
IP
H + ?ZP
75
1 and 2; and F,y are the flexibilities indicated in Fig. 5. These flexibilities
can be considered deterministic based on the observations for beams and
single-story frames. The statically equivalent wind forces, H,, are com-
puted from an analysis that takes into account the spatial coherence and
temporal variations in the fluctuating wind pressure. Thus, the forces
Hi can be assumed to be perfectly correlated, i.e., they are either equal
or they are proportional to each other. In general, it can be seen that
for any mutli-story frame
Pi = H(F,) (13)
where H is a reference lateral force; and F, is the appropriate frame flex-
ibility which now also includes the deterministic proportionality factors.
For example, for a two story fame under loads Ha = H and H2 = 0.5H,
Pi = H(Fn + 0.5F12) and p2 = H(F21 + 0.5F22).
The form of Eq. 13 is thus the same as that of Eq. 10 for a single story
frame, and so the reliability index is expressed by Eq. 11, regardless of
the number of stories. This is a consequence of the perfect correlation
of the statically equivalent lateral loads. The reliability index for lateral
drift in multi-story frames due to windloads is thus approximately 1.5
for an eight-yr, and 3.2 for a one-yr reference period, respectively.
The reliability indices p for the deflection and lateral drift serviceability
limits described previously can be related, to an order of magnitude, to
a serviceability limit state probability, pf, by
P/=*(-P). (14)
in which <3>( ) is the standard normal probability integral. Substituting
in the value of (J listed in Tables 3 and 4 the probability that a structure
designed to satisfy current deflection or drift guidelines will experience
a serviceability problem of the type described herein is on the order of
5% over an eight-yr period and 0.1% or less on an annual basis.
The reliability index associated with current design practice, using first-
order probabilistic methods is examined. The deflection limits are taken
to be the customary deterministic values which are assumed to provide
protection against unsightly deflections and against damage to nonstruc-
tural elements and equipment. A reference period of eight yrs was as-
sumed for the reliability analyses which represents the average period
of one tenancy in office buildings. Reliability analyses were also made
for a one-yr period, applicable to situations where deflection perception
is not critical or where cladding and equipment is not sensitive to de-
flection.
The following values of the reliability index p were obtained in this
analysis for beams and frames where restraint effects were not included
(the P's are for an eight-yr reference period interval):
ACKNOWLEDGMENT
The financial support of the American Iron and Steel Institute is grate-
Downloaded from ascelibrary.org by University Of Sydney on 09/20/13. Copyright ASCE. For personal use only; all rights reserved.
fully acknowledged.
0.4
0.90 < <0.99
H
+ 0.5
kL
if the higher spring constant is matched with the stiffer beams. In the
following analysis it will be assumed that the actual deflection may vary
between 80% and 99% of the "simple" beam deflection.
If we denote the flexibility F as
5L4 0.4
F= (16)
384 EI EI
— + 0.5
kL
5L3
and let C= (17)
384
be a deterministic variable while E, I, and k are assigned to be indepen-
dent random variables with mean values E, I and k and COV's VE, Vt
and Vk, then the mean flexibility and its COV are equal to, with S =
EI/L
78
0.4S 0.4S
Downloaded from ascelibrary.org by University Of Sydney on 09/20/13. Copyright ASCE. For personal use only; all rights reserved.
0.4
-1 + S
-. + 0,5 ~. + 0.5 + 0.5
k
VF = - (19)
0.4
1 -
S
-= + 0.5
k
Taking the values E = En, VE = 0.06, I = I„, V, = 0.05 (4), w h e r e the
subscript n denotes nominal quantities, then
El
S= — and Vs = Vo.06 2 + 0.05 2 = 0.08
L
If a relatively high COV of Vk = 0.3 is taken for the variability of the
spring constant, then Eq. 19 gives Vr = 0.08 for the practical range of
S/k = 700,000/20,000 and 20,000/5,000. The variability of the end-re-
straint thus has no effect on the COV of the flexibility, and t h u s VF =
0.08 will be used in the following derivations.
In the reliability analysis it is a s s u m e d that the allowable deflection 8„
is a deterministic limit specified by the designer or by a building code.
The limit state equation is
£=5fl-F*L* = 0 (20)
where L* and F * are the "checking p o i n t " values (4) of the load a n d the
flexibility, respectively.
F* = F(l-aFVF\S) (21)
L* = L ( l - a L y L ( 3 ) (22)
The terms in these equations are: F, Vr and L, VL are the m e a n s a n d the
COV's of the flexibility a n d the load, respectively; p is the reliability
index (4), and
af
dF*
a.F = (23)
D
CTL
dV
<*!. = • (24)
D
where CTF = VFF and a t + VLL are the standard deviations of F a n d L,
and
79
-F*<JL
aL = . (27)
2
VL* <TF + F * V
The contemporary design requirement for checking floor or roof de-
flection is
where the subscripts n denote nominal values. At the limit F„Ln = 8„,
Eq. 28 gives
C8„
Ln = ^ (29)
The mean load is given by
£ L (3o)
=(£) "
where L/L„ is given in Table 1 for occupancy and snow loads. Substi-
tution of Eq. 29 into Eq. 30 gives
"©(?) -
and therefore also
v £ (32
*- ' -"'©(?) : >
Introducing (Eqs. 21 and 22)
F, = -j = 1 - aFVF , (33)
U = j = 1 - aLVL (34)
0.4
and Y=1- (35)
- + 0.5
k
we obtain the following relationships:
i=G=1-{BmLi • (36)
80
«L = 5 (38)
Downloaded from ascelibrary.org by University Of Sydney on 09/20/13. Copyright ASCE. For personal use only; all rights reserved.
, 2 , \ 2
,39)
"•^"yid
The determination of p is made in the following manner:
Table 2 gives the results of such a calculation for Y varying from 0.8-
0.99.
F =^ + ~JL-L (41)
6 EL. 12 EL,
Let Ic = I, the reference moment of inertia; then
1 Ec ECEB-/C
F=— (42)
EI 6 12Z B
The bracketed term in this equation can be considered to be determin-
istic, and thus the COV of the flexibility is
VF = W l + Vj (43)
Since VE = 0.06 and Vl = 0.05 (4), VF = 0.08.
The limit state equation is
g = Pa - F*H* = 0 (44)
81
shown that
H„ (45)
VH
if the variability of the flexibility F is considered negligible (i.e., Vr = 0)
in comparison to the variability of the loads. In this case, p = 1.6 and
3.4, respectively, for the eight-yr and the one-yr reference periods.
Identical results are obtained for X-braced and K-braced simple frames
(Fig. 3), for which, respectively
1 Abr
F= + (46)
EAbr .LCAB cos*-r
t]L, ^L Ac
1 1
I (hi (47)
EAbr 2\LC A •'£
vsm i\ cos r\/
P-delta Effect.—In the presence of gravity load (Fig. 4) the lateral force
is (2) H + P 2P, and thus
p = (H+ p2P)F . (48)
HF
from which p= (49)
1 -FI.P
The limit state equation then is equal to
FH*
Pa = 0 (50)
1 -F2P*
if it is assumed that in comparison with the loads the variability of the
flexibility can be neglected
H* = H(l - a H l/„p) (51)
and 2P* = 2P(1 - a 2P V SP P) (52)
The gravity load 2P is made up of possible contributions from dead
occupancy live or snow loads, and it can be shown that
EP = 2PDN[pD + lpL + sps] (53)
PD PL
# , = - £ = 1.05, VD = 0.1(4); ft = —
1 Drc L Ln
H_
Pw (56)
p
The values of pL, ps, and pw, as well as the corresponding COV's are
listed in Table 1.
Pa,. -3L
/ = • (57)
PD,,'
"-PD,,
VXP
aH : __ . aj (58)
d2PD
D = + °"1P (59)
tff' ffH 32P
where crH = VH/H and o-2P = V 2P /XP.
If the terms
X = w + p„(l + / + s); Z = pD + lpL + spsi
Y1 = l-aHVHfr Y2 = l - a E P ^ 2 P p (60)
are introduced, then
(61)
Pa X - PaZY2
and , _ ^ H = - f c | _ _ (62)
APPENDIX III.—REFERENCES
83
APPENDIX IV.—NOTATION
A = cross-sectional area,
D = dead load,
E = modulus of elasticity,
F = flexibility,
g = checking equation,
h = story height,
J = m o m e n t of inertia,
k = spring constant,
L = span,
L = live load,
P = gravity load,
S = snow load,
V = coefficient of variation,
W = wind load,
w = distributed load,
a = coefficient,
P = reliability index,
A = deflection,
8„ = allowable deflection,
T| = brace inclination, and
p = drift.
84