Professional Documents
Culture Documents
Paper 2
Paper 2
Axial-Load Effects
Keri L. Ryan, M.ASCE1; James M. Kelly, M.ASCE2; and Anil K. Chopra, M.ASCE3
Abstract: Existing models for isolation bearings neglect certain aspects of their response behavior. For instance, rubber bearings have
been observed to decrease in stiffness with increasing axial load, and soften in the vertical direction at large lateral deformations. The yield
Downloaded from ascelibrary.org by Iowa State University on 09/27/13. Copyright ASCE. For personal use only; all rights reserved.
strength of lead–rubber bearings has also been observed to vary with axial load, such that a lightly loaded bearing may not achieve its
theoretical strength. Models that include these axial-load effects in lead–rubber bearings are developed by extending an existing linear
two-spring model to include nonlinear behavior. The nonlinearity includes an empirical equation for the experimentally observed variation
of yield strength. For numerical implementation, the bearing forces are found by solving the nonlinear equilibrium and kinematic
equations using Newton’s method, and the instantaneous bearing stiffness matrix is formed from the differentials of these equations. The
response behavior of the models is confirmed by comparison with experimental data.
DOI: 10.1061/共ASCE兲0733-9399共2005兲131:12共1270兲
CE Database subject headings: Axial forces; Axial loads; Rubber; Lead; Base isolation; Numerical models; Nonlinear analysis;
Plasticity.
Stability Analysis of Multilayer Bearings where Ec = instantaneous compression modulus of the rubber–
steel composite bearing. In this context, the term nominal 共de-
In the stability analysis of multilayer bearings 共Koh and Kelly noted by subscript o兲 means absent axial-load effects.
1987; Kelly 1997兲, the bearing is treated as a continuous compos- If the shear stiffness of the two-spring model 共Fig. 2兲
ite system in which the steel layers do not deform, allowing pre- were infinite, the rotational stiffness divided by hb would
diction of the buckling load and the effective lateral stiffness in equal the conventional Euler buckling load PE = 共2 / h2b兲EIs. Here
the presence of axial load. The stability theory resembles the lin- EIs⫽ bending stiffness of a multilayer bearing 共Kelly 1997兲
earized theory of an elastic column, but accounts for shear defor-
1 hb
mation by considering rotation of the cross section, which is in- EIs = EcI 共3兲
dependent of the lateral deflection 共Koh and Kelly 1987兲. Also 3 tr
predicted by stability analysis 共Koh and Kelly 1987兲, the where I = conventional moment of inertia: D4 / 64 or Ar2b;
multilayer bearing under simultaneous lateral and axial loading and rb = D / 4 = bending radius of gyration in terms of the bearing
undergoes an additional vertical displacement beyond that due to diameter D. Thus, the rotational stiffness, divided equally among
material axial flexibility. Demonstrated visually in Fig. 1, the ad- top and bottom springs 共Fig. 2兲, is PEhb.
ditional displacement ␦bz, either compressive or tensile depending With these simple linear constitutive relations, the equilibrium
on the corresponding axial load, is due to tilting of the middle equations relating the lateral force f b and axial 共compressive兲
bearing layers when the bearing is deformed in shear. force P to the deformation s across the shear spring and the ro-
tation through the rotational spring 共Fig. 2兲 are
Approximate Force–Deformation Relation Based
f b − kbos + P = 0 共4a兲
on Linear Two-Spring Model
The preceding effects predicted by stability analysis can be rep- f bhb − PEhb + P共s + hb兲 = 0 共4b兲
resented, approximately, by a simplified two-spring model of the
bearing 共Kelly 1997兲 that results in explicit force–deformation which assume small rotation . The axial force P and the defor-
relations. The two-spring model 共Fig. 2兲 is a composition of rigid mation v across the vertical spring 共not shown in Fig. 2兲 are
tees connected by a rotational spring, subdivided at top and bot- related by an additional equation
tom, and a shear spring with frictionless rollers at midheight. The P − kbzov = 0 共4c兲
bottom plate is fixed and the top plate is constrained against ro-
tation. Axial flexibility of the bearing is included by an additional The kinematic equations relating the total lateral deformation
vertical spring in series 共not shown in Fig. 2兲. ub and vertical deformation ubz to the internal deformations s, ,
Assuming linear material behavior, the nominal shear stiffness and v, again assuming to be small, are
of a multilayer bearing, also the stiffness of the shear spring in
Fig. 2, is u b = s + h b 共5a兲
GA GAs hb 2
kbo = = 共1兲 ubz = v + ␦bz = v + s + 共5b兲
tr hb 2
where G = shear modulus; A = cross-sectional area; and tr = sum In Eq. 共5b兲, ubz, positive in compression, is the sum of v, the
thickness of the rubber layers. In some cases, it is convenient to deformation resulting from axial flexibility of the bearing, and
use modified area As = A共hb / tr兲, where hb⫽total height of the bear- ␦bz, the additional vertical displacement that occurs in the later-
ing, which accounts for the undeforming steel layers. Similarly, ally deformed configuration as shown in Figs. 1 and 2.
substituted into Eq. 共5a兲, which, after approximation, simpli- Lateral and axial forces are normalized by static, or gravity load Pst.
fies to the following lateral force–deformation relation:
f b = k bu b where 冋 冉 冊册
kb = kbo 1 −
P
Pcr
2
共7兲
Written in flexibility form 关Eq. 共8兲兴, the vertical force–
deformation relation shows two distinct contributions to the ver-
tical deformation: a material effect and a second order geometric
Again, the lateral stiffness kb in Eq. 共7兲 for the two-spring effect. The additional vertical displacement ␦bz due to geometry
model is a good approximation to the stiffness derived from of deformation 关second term of Eq. 共8兲兴 increases according to the
stability analysis of the multilayer bearing 关Kelly 共1997兲, square of the lateral deformation. Note that even with no axial
Fig. 8-4兴. load on the bearing, the additional displacement ␦bz is nonzero,
3. These values of s and , as well as the axial deformation v albeit small because PE Ⰷ PS, as shown in Fig. 3共b兲 at P = 0 for
determined from Eq. 共4c兲, are substituted into Eq. 共5b兲. This different lateral deformations ub. The net effect of this additional
leads to the following flexibility equation for vertical defor- displacement is an overall softening of the bearing in the vertical
mation as a function of force: direction, which depends on the lateral deformation 关Eq. 共10兲兴.
P 共PS + P兲 u2b This softening is also evident in the vertical force–deformation as
ubz = + 共8兲 a function of lateral deformation ub 关Fig. 3共b兲兴, wherein the slope
kbzo PE hb of the line for ub = 0 represents the nominal vertical stiffness 关Eq.
which is inverted to give 共2兲兴. Physically, the softening occurs as a result of the tilting of
bearing reinforcing layers 共Fig. 1兲, meaning the axial loads are
P = kbz ubz −冉 PS u2b
PE hb
冊 共9兲
resisted in part by shear.
with incremental, or tangent vertical stiffness kbz given by Variation of Yield Strength with Axial Load
kbz = 冉 1
+
kbzo PEhb
u2b
冊 −1
冉
= kbzo 1 +
2r2b
冊
3u2b −1
共10兲 Experimental evidence 共Tyler and Robinson 1984; Hwang and
Hsu 2000兲 of the bearing failing to achieve its full strength when
reduced to its final form by substituting Eq. 共3兲 for PE. Thus, lightly loaded has been attributed to lack of confinement of the
the lateral and vertical force–deformation relations of the lead plug. Skinner et al. 共1993兲 writes “the nominal upper limit of
bearing are defined by the coupled Eqs. 共7兲 and 共9兲. If axial- hysteretic force … should be achieved if there is no vertical slip-
load effects are neglected 关i.e., axial load P = 0 in Eq. 共7兲 and page of the plug sides and no horizontal slippage of the plug
lateral deformation ub = 0 in Eq. 共9兲兴, these equations reduce ends.” Current methods known to achieve the best confinement
to the familiar uncoupled linear force–deformation equations include minimizing the thickness of the rubber layers, fitting an
with nominal stiffnesses kbo 关Eq. 共1兲兴 and kbzo 关Eq. 共2兲兴. oversized lead plug into the undeformed cavity, and using a spe-
These stiffnesses are slightly modified in Ryan and Chopra cial technique to cap the plug. However, even recently tested
共2005兲 to include the effect of bulk compressibility. lead–rubber bearings seem to have this deficiency 共Hwang and
Lateral force–deformation curves for different axial loads are Hsu 2000兲.
shown in Fig. 3共a兲, with the curve corresponding to no axial load Because the preceding observation has not been justified by
共P = 0兲 representing the nominal stiffness. Each curve is linear mechanical theory, we have developed an empirical equation for
because P is constant. Together, the set of curves indicate the the yield strength as a function of the compressive load P based
reduction in lateral stiffness as the axial load increases. The stiff- on experimental data:
ness variation is inconsequential for typical design values of
Q = Qo共1 − e共−P/Po兲兲 共11兲
P / Pcr ⬍ 0.2, but the stiffness rapidly approaches zero as the axial
load approaches the critical load. where Qo = nominal yield strength of the bearing, achievable with
Consideration of the negative solution to Pcr 关Eq. 共6兲兴 leads to an adequate confining pressure; and the axial load Po, correspond-
a critical load in tension, hence the unconventional concept of ing to about 63% of nominal strength, should be chosen to match
tension buckling 共Kelly 2003兲. Because the two solutions to Pcr characteristic test data for the bearings. In Fig. 4, a plot of Eq.
are equal and of opposite sign, the bearing buckling behavior in 共11兲 shows that the strength declines quite rapidly for loads P
tension mirrors that in compression. Thus, the stiffness reduction below Po. The nominal strength is only achieved 共at 95% or
of Eq. 共7兲 is the same for both compressive and tensile loads greater level兲 when P 艌 3Po, therefore we recommend a minimum
共positive and negative values of P兲. of 3Po for the bearing design load Pst. When the bearing is in
Numerical Implementation
An approach is presented to integrate the constant-strength and
Fig. 7. Lateral force–deformation as function of P / Pcr for: variable-strength bearing models into a typical dynamic analysis
共a兲 constant-strength and 共b兲 variable-strength models program. This approach is compatible with programs that com-
Downloaded from ascelibrary.org by Iowa State University on 09/27/13. Copyright ASCE. For personal use only; all rights reserved.
冦冧冦 冧
f b − f s共s兲 + P = 0 共12兲 g1 f b − f s共s兲 + P
where the linear shear spring has been replaced by a general force g2 f bhb − PEhb + P共s + hb兲
f s共s兲.
g= g3 = P − kbzov 共15兲
The force of the shear spring in the constant-strength model
may be determined numerically by classical rate-independent uni- g4 u b − s − h b
directional plasticity 共Simo and Hughes 1998兲, where the force– g5 ubz − s − hb2/2 − v
deformation relation is elastic–plastic with kinematic hardening,
and solved by Newton’s method, i.e., find x = 具f b , P , s , , v典T to
and thus governed by the following constitutive law, yield func-
satisfy g共x兲 = 0.
tion, flow rule, and hardening law:
The converged solution x at the previous global iteration
f s = kI共s − s p兲 共13a兲 serves as an initial guess x共0兲. An improved solution is found by
冤 冥
q̇ = ␥˙ H sgn共f s − q兲 共13d兲 fs
1 − P 0
s
respectively. The constitutive law 关Eq. 共13a兲兴 shows that the
spring force equals the initial stiffness kI times the elastic compo- hb 共s + hb兲 P 共P − PE兲hb 0
J共x兲 = 共17兲
nent of deformation s − s p, s p being the plastic deformation. The 0 1 0 0 − kbzo
initial stiffness kI = kbo + Qo / sy depends on the nominal stiffness 0 0 −1 −h 0
and strength, as well as the yield deformation sy. The yield func-
0 0 − − 共s + hb兲 − 1
tion ⌽ 关Eq. 共13b兲兴 determines the set of admissible forces, where
the back force q stores the translation of the yield surface, and the has a nonzero determinant, and hence is invertible, even when P,
yield force f y = Qo + kbosy. The spring response is elastic inside the s, and are zero. Eq. 共16兲 is applied repeatedly until the incre-
yield surface 共⌽ ⬍ 0兲, and plastic flow occurs on the yield surface mental change in the solution is less than a desired tolerance, that
共⌽ = 0兲, determined by the flow rule 关Eq. 共13c兲兴 with constant slip is: 储x共k兲 − x共k−1兲储 ⬍ tol. At each iteration the shear spring force f s共s兲
rate ␥˙ . Evolution of the back force is governed by the hardening in Eq. 共15兲 and tangent f s / s in Eq. 共17兲 are computed by the
law 关Eq. 共13d兲兴, with hardening stiffness H = kIkbo / 共kI − kbo兲. return mapping algorithm 共Ryan and Chopra 2005兲.
冤 冥
1
dU = 具dub , dubz典T, is derived in three steps:
1. Take differentials of the equilibrium equations 关Eqs. 共12兲, T = hb 共s + hb兲 共19b兲
共4b兲, and 共4c兲兴, resulting in 0 1
2. Take differentials of the kinematic equations 关Eq. 共5兲兴, result-
keqdv = TdF 共18兲 ing in
dU = TTdv 共20兲
where dv = 具ds , d , dv典T and the matrices keq and T are given
by 3. Substitute dv from Eq. 共18兲 into Eq. 共20兲
Downloaded from ascelibrary.org by Iowa State University on 09/27/13. Copyright ASCE. For personal use only; all rights reserved.
冤 冥
fs dU = 共TTfeqT兲dF 共21兲
−P 0
s −1
keq = 共19a兲 where feq = keq . The resultant flexibility matrix fb = TTfeqT,
− P 共PE − P兲hb 0 which relates the incremental deformations dU to the incre-
0 0 kbzo mental forces dF, is given explicitly as
冤 冥
共PE + P兲hb + h2b共 f s/s兲 PEhb + 关P + hb共 f s/s兲兴共s + hb兲
hb共 f s/s兲共PE − P兲 − P 2
hb共 f s/s兲共PE − P兲 − P2
fb = 共22兲
PEhb + 共P + hb f s/s兲共s + hb兲 共PE + P兲hb2 + 2Ps + 共 f s/s兲共s + hb兲2 1
+
hb共 f s/s兲共PE − P兲 − P2 hb共 f s/s兲共PE − P兲 − P2 kbzo
1. Suppose both the axial load P and deformations s and are which is identical to the approximate lateral stiffness kb in
zero, and the shear spring is linear 共 f s / s = Ps / hb兲; then the Eq. 共7兲.
flexibility matrix 关Eq. 共22兲兴 is diagonal with elements 3. Suppose hb is small relative to the total lateral deformation
ub, the shear spring is linear, and P Ⰶ Pcr; then f 22 can be
共PE + PS兲hb approximated as
f 11 = 共23a兲
PE PS
1 u2b 1
and f 22 ⬇ + 共26兲
PE hb kbzo
when the axial force is close to zero by comparing Figs. 10共b兲 and
11共b兲. Observe that the strength decreases and the yield surface
contracts when the lateral force is negative 关Fig. 11共b兲兴. Com-
pared to the uncoupled nonlinear model, the average width of the
force–deformation loop is smaller, and the associated decrease in
energy dissipation causes the peak lateral deformation to increase,
Fig. 11. Response of left exterior bearing—using variable-strength which in turn increases the peak vertical deformation 关Fig. 11共c兲兴.
model—of rigid block subjected to seismic pulse: 共a兲 lateral and
The local oscillations in the lateral force–deformation are stronger
vertical deformation and force histories 共qualitative, with relative
compared to Fig. 10共b兲 as the behavior is affected by local
scale indicated兲, 共b兲 lateral force–deformation, and 共c兲 vertical
strength variations as well as stiffness variations.
force–deformation. Also shown is comparative response with
uncoupled nonlinear model; Pst / Pcr = 0.4 and Pst / Po = 3.
Conclusions
load effects are included 关coupled linear model versus uncoupled
linear model, Figs. 9共a and c兲兴. To understand this, recall that the This investigation to develop a nonlinear model for isolation bear-
vertical force–deformation relation depends greatly on the lateral ings that includes the influence of axial load has led to the fol-
deformation 关Eq. 共9兲兴. Thus, when the axial force is close to lowing conclusions:
its static value, the lateral deformation is close to zero 共recalling 1. From testing of both high-damping rubber and lead–rubber
the correlation between P and ub observed above兲 and the vertical bearings, it has been observed that the lateral stiffness de-
force–deformation is essentially that of the uncoupled linear creases with increasing axial load, the lateral yield strength
model. However, as the axial force deviates from the static decreases with decreasing axial load 共lead–rubber bearings
force in either direction, the lateral deformation becomes large, only兲, and the vertical stiffness decreases with increasing lat-
resulting in vertical softening and a considerable increase in ver- eral deformation.
tical deformation, and thus the arc-shaped vertical force– 2. Based on linear stability theory of an elastic column, a two-
deformation 关Fig. 9共c兲兴, with vertical deformation increasing at spring model of the bearing, consisting of a shear spring and
axial forces larger or smaller than Pst. Due to the inherently linear a rotational spring divided at top and bottom 共Fig. 2兲, accu-
relation between axial force P and deformation ubz 关Eq. 共9兲兴, rately accounts for axial-load effects in linear models of
the total deformation ubz is obviously greater when P ⬎ Pst than bearings. However, bearings with high-damping fillers or
when P ⬍ Pst. lead cores that provide energy dissipation are nonlinear and
The bearing response predicted by the constant-strength model should be modeled as such.
is plotted in Fig. 10 and compared with the response predicted by 3. Such modeling is achieved by extending the two-spring
the uncoupled nonlinear model. Whereas for the linear models the model to include a nonlinear constitutive model for the shear
axial force history and lateral deformation history were in phase spring. To form the constant-strength bearing model, this
关Fig. 9共a兲兴, for these nonlinear models the axial force history is shear spring was modeled by unidirectional plasticity with
more closely in phase with the lateral force history 关Fig. 10共a兲兴. kinematic hardening. Numerically, the model is implemented
For the constant-strength model, this causes the lateral stiffness by viewing the equilibrium 关Eqs. 共12兲, 共4b兲, and 共4c兲兴 and
共initial or postyield兲 to show the greatest decrease at large positive kinematic 关Eq. 共5兲兴 equations as a system of five nonlinear
lateral forces 关Fig. 10共b兲兴, which correspond to the maximum equations, which are solved by Newton’s method for the cur-
axial forces. The result is a slight increase in peak lateral defor- rent forces in the bearing. The instantaneous stiffness matrix
mation compared to the uncoupled nonlinear model. Because of is derived by taking differentials of these equations. Al-
the changing stiffness, the apparent width of the hysteresis loop though an improvement, the model is an incomplete repre-
changes during the deformation cycle, whereas this width is con- sentation of high-damping rubber bearings because it ne-
stant for the uncoupled nonlinear model. A high-frequency com- glects complex effects like strain hardening.
ponent is observed in the axial force cycle 关Fig. 10共a兲兴, causing 4. An empirical model 关Eq. 共11兲兴 was developed to account for
local stiffness variations and small oscillations in the lateral the varying yield strength in lead–rubber bearings, which can
force–deformation relation 关Fig. 10共b兲兴. be calibrated to match the experimentally observed bearing
The vertical force–deformation relation of the constant- response. This behavior is optionally included by simply up-
strength model is difficult to interpret 关Fig. 10共c兲兴. Although quite dating the yield force in the plasticity equations to reflect the
The writers are grateful for the funding of this research contrib- Earthquake Engineering Simulation 具http://opensees.berkeley.edu,
uted by a California state grant. 2004典.
Ryan, K. L., and Chopra, A. K. 共2005兲. “Estimating the seismic response
of base-isolated buildings including torsion, rocking, and axial-load
References effects.” Rep. No. UCB/EERC-2005/01, Earthquake Engineering Re-
search Center, Univ. of California, Berkeley, Calif.
Aiken, I. D., Kelly, J. M., and Tajirian, F. F. 共1989兲. “Mechanics of low Ryan, K. L., Kelly, J. M., and Chopra, A. K. 共2004兲. “Experimental
shape factor elastomeric seismic isolation bearings.” Rep. No. UCB/ observation of axial-load effects in isolation bearings.” Proc., 13th
EERC-89/13, Earthquake Engineering Research Center, Univ. of Cali- World Conf. on Earthquake Engineering, Paper No. 1707, Canadian
fornia, Berkeley, Calif. Association for Earthquake Engineering, Vancouver, British Colum-
Clark, P. W., Aiken, I. D., and Kelly, J. M. 共1997兲. “Experimental studies bia, Canada.
of the ultimate behavior of seismically-isolated structures.” Rep. No. Simo, J. C., and Hughes, T. J. R. 共1998兲. Computational inelasticity,
UCB/EERC-97/18, Earthquake Engineering Research Center, Univ. of Springer, New York.
California, Berkeley, Calif. Skinner, R. I., Robinson, W. H., and McVerry, G. H. 共1993兲. An intro-
Griffith, M. C., Kelly, J. M., Coveney, V. A., and Koh, C.-G. 共1988兲. duction to seismic isolation, Wiley, New York.
“Experimental evaluation of medium-rise structures subject to uplift.” Tyler, R. G., and Robinson, W. H. 共1984兲. “High-strain tests on lead-
Rep. No. UCB/EERC-88/02, Earthquake Engineering Research Cen- rubber bearings for earthquake loadings.” Bull. New Zealand Nat.
ter, Univ. of California, Berkeley, Calif. Soc. Earthquake Eng., 17共2兲, 90–105.