Professional Documents
Culture Documents
Abstract
The responses of multistory flexibly connected frames subjected to earthquake excitations are studied using a computer model.
The model incorporates connection flexibility as well as geometrical and material nonlinearities in the analyses. Connections are
modeled as rotational springs with bilinear hysteretic moment–rotation relationships. Geometrical nonlinearities in the form of
member and frame P-delta effects are considered by the use of stability functions in a stiffness formulation. Material nonlinearity
in the form of column inelasticity is accounted for by the use of the tangent modulus concept. Response characteristics of two
multistory frames with three connection types (rigid, semirigid and flexible) subjected to two earthquakes are studied with reference
to their modal response attributes. The study indicates that connection flexibility tends to increase upper stories’ interstory drifts
but reduce base shears and base overturning moments for multistory frames. Connection flexibility also causes the frame periods
to spread over a wider spectrum and increases the importance of higher mode contributions to structural response. When the
Complete Quadratic Combination (CQC) rule of spectral analysis is applied to flexibly connected frames, it is found that noticeable
errors may be produced. 1998 Elsevier Science Ltd. All rights reserved.
Keywords: Flexibly connected frames; Modal responses; Seismic analysis; Semirigid connections
0141-0296/99/$—see front matter 1998 Elsevier Science Ltd. All rights reserved.
PII: S 0 1 4 1 - 0 2 9 6 ( 9 7 ) 0 0 2 1 0 - 1
426 J.C. Awkar, E.M. Lui / Engineering Structures 21 (1999) 425–441
connected frames have also been shown [10] to be more 2. Structure model
sensitive to P-delta effect. Experimental studies [11–15]
on flexibly connected frames have demonstrated that The structures used in the present study are single bay
properly designed semirigid connections can exhibit multistory flexibly connected frames. The structure stiff-
reliable hysterectic responses, thus making flexibly con- ness matrix is obtained by assembling column and beam
nected frames suitable candidates for use in seismic element stiffness matrices to be discussed in the follow-
application. ing sections. The effects of member/frame instability
In this paper, the results of a numerical study on the (i.e. the P-delta effects) and column inelasticity are
responses of multistory flexibly connected frames sub- incorporated in the column stiffness matrix through the
jected to earthquake excitations will be presented. A use of stability functions and the tangent modulus con-
modal analysis procedure will be used for the study, and cept, respectively. The effect of connection flexibility is
the responses of the frames will be assessed by examin- incorporated in the beam stiffness matrix through the use
ing their modal responses. In a modal analysis, the total of spring elements located at the beam’s ends.
response of a system is obtained by a linear combination
of the system modal responses. Modal analysis will yield 2.1. Column model
accurate results if the system behaves in a linearly elastic
fashion and when sufficient number of modes are The column model is shown in Fig. 1. It is a 4-DOF
included in the analysis. While modal analysis has suc- prismatic beam–column element. The stiffness matrix
cessfully been applied to obtain responses for classically that relates the nodal displacements dc = {d1d2d3d4}T to
damped multidegree of freedom (MDOF) linear systems the nodal forces rc = {r1r2r3r4}T is given by [17]
under various forms of excitations, the applicability of
the method to evaluating the seismic responses of multi- 12 6 12 6
story flexibly connected steel frames, which often exhibit 2
⌽1 − ⌽2 − 2 ⌽1 − ⌽2
Lc Lc Lc Lc
nonlinear behavior, has not been demonstrated. In the
6
present study, a modified form of the modal analysis pro- EIc 4⌽3 ⌽ 2⌽4
cedure is introduced for evaluating the seismic responses Kc = Lc 2 (1)
Lc
of two multistory frames considering connection flexi- 12 6
bility. The connections are modeled as rotational springs ⌽ ⌽
L2c 1 Lc 2
that exhibit a bilinear moment–rotation behavior. The
flexibility effect of the connection is incorporated in the Sym. 4⌽3
analysis by introducing a fictitious force vector to the
equations of motion. where E is the elastic modulus, Ic and Lc are the moment
In addition to accounting for connection flexibility, the of inertia and length of the column, respectively. ⌽1, ⌽2,
⌽3, and ⌽4 are stability functions that account for the
analysis takes into consideration geometrical and
presence of an axial force (and thus the P-delta effects)
material nonlinearities. Geometrical nonlinearities in the
in the member. For a compressive axial force P these
form of P-delta effects are accounted for by the use of
functions are given by
stability functions in a stiffness formulation, and material
nonlinearity in the form of column inelasticity is con-
sidered through the use of the tangent modulus concept.
A Mathcad [16] computer program has been
developed to perform the aforementioned modal analy-
sis. The program generates the responses in all modes for
different response quantities (roof displacements, base
shear, base overturning moment, etc.). One objective of
the present study is to determine the minimum number
of modes that is needed to yield accurate responses for
flexibly connected frames. Another objective of the
study is to investigate the degree to which higher modes
contribute to the responses. A third objective is to inves-
tigate the applicability of the Complete Quadratic Com-
bination (CQC) rule of spectral analysis to flexibly con-
nected frames.
(kLc)3sinkLc elements. The stiffness matrix that relates the nodal dis-
⌽1 = (2) placements db = {d1d2}T to the nodal forces rb = {r1r2}T
12⌽c
is given by [17]
(kLc)2(1 − coskLc)
冋 册
⌽2 = (3)
6⌽c EIb s*ii s*ij
Kb = (9)
Lb s*ij s*jj
kLc(sinkLc − kLccoskLc)
⌽3 = (4)
4⌽c
where
kLc(kLc − sinkLc)
⌽4 = (5) 12EIb
2⌽c 4+
LbRBk
s*ii =
冉 冊冉 冊 冉 冊
in which 2
(10)
4EIb 4EIb EIb 4
1+ 1+ −
⌽c = 2 − 2coskLc − kLcsinkLc LbRAk B A B
(6) LbRk Lb Rk Rk
冪EI
P 2
s*ij =
冉 冊冉 冊 冉 冊
k= (7) 2
(11)
c 4EIb 4EIb EIb 4
1+ 1+ −
LbRAk LbRBk Lb RAkRBk
The stability functions take a slightly different form if
the column axial force is tensile. However, these func-
tions are not shown here because no column in the 12EIb
4+
present study experiences a tensile axial force. LbRAk
s*jj =
冉 冊冉 冊 冉 冊
(12)
The quantity in Eqs. (1) and (7) accounts for inelas- 4EIb 4EIb EIb 2
4
ticity in the column. It is taken as the ratio of the tangent 1+ 1+ −
LbRAk LbRkB
Lb A B
Rk Rk
modulus Et to the elastic modulus E obtained by dividing
the AISC-LRFD [18] inelastic column curve by the elas- where E is the elastic modulus, Ib and Lb are the moment
tic column curve of inertia and length of the beam, respectively. RAk and
RBk are the instantaneous stiffness of the connections at
冦
P the A and B ends of the beam, respectively. Values of
1, if ⱕ 0.39
Et Py RAk and RBk are to be obtained from the hysterectic con-
= =
冉冊
(8) nection model to be discussed in the following section.
E P P P
− 2.724 ln , if > 0.39 For the present study, axial deformations in the col-
Py Py Py
umns and beams are neglected because these defor-
mations are small compared to the bending defor-
where Py is the yield load of the column. The tangent mations. As a result, the column model consists of only
modulus concept is a simple method to account for lateral translational and rotational degrees of freedom,
yielding in the column. The method works well, pro- and the beam model consists of only rotational degrees
vided that no plastic hinge forms in the column. From a of freedom.
physical standpoint, the term EIc represents the inelastic
flexural rigidity of the column that is available to carry
2.3. Connection model
moment generated by any lateral load, following the full
application of the gravity loads.
Despite the usual assumption of fully rigid or ideally
2.2. Beam model pinned, connections in reality are neither rigid nor
pinned. All connections experience some types of defor-
The beam model is shown in Fig. 2. It is a hybrid mations when subjected to forces and moments. In gen-
element consisting of a flexural element and two spring eral, most connections can transfer axial forces, shears
and moments. However, for practical purposes only
bending deformation is considered because it is the pre-
dominant deformation for ordinary beam-to-column con-
nections.
Connection moment–rotation curves that relate bend-
ing moments to rotations are usually obtained from
experiments. Numerous curves have been generated and
documented [19–21]. These curves are nonlinear from
Fig. 2. Two degrees of freedom beam model. the beginning of the loading to the end. In order to incor-
428 J.C. Awkar, E.M. Lui / Engineering Structures 21 (1999) 425–441
冋 册 冋 册冋 册
is equal to Rkp and it intersects the ordinate at Mbound.
F(t) ktt ktr ut
= (13)
2.4. Structure stiffness and static condensation 0 krt krr ur
The structure stiffness matrix is obtained by assemb- where ut is an N ⫻ 1 vector containing the translational
ling the column and beam stiffness matrices described DOFs and ur is a 2N ⫻ 1 vector containing the rotational
above according to conventional stiffness matrix analy- DOFs; one obtains after expanding Eq. (13) the follow-
sis procedure. For an N-story one-bay frame, the system ing two equations:
stiffness matrix so assembled is of size 3N ⫻ 3N as
shown in Fig. 4. F(t) = kttut + ktrur (14)
To simplify the analysis, the number of degrees of
freedom (DOF) of the frame is to be reduced by stati- 0 = krtut + krrur (15)
cally condensing out the rotational DOFs and retaining
only the translational DOFs. This can be done readily in Solving Eq. (15) for ur and substituting the result into
the context of a dynamic analysis if a lumped mass Eq. (14), one obtains
model is used and the inertia forces associated with the
rotational DOFs are neglected. With reference to Fig. 4, F(t) = kut (16)
if the translational DOF and rotational DOF are separ-
ated and partitioned as follows where
k = ktt − ktrk−1
rr krt (17)
3. Method of analysis
Fig. 3. Bilinear hysterectic connection model. in which m, c, and k are the N ⫻ N mass, damping and
J.C. Awkar, E.M. Lui / Engineering Structures 21 (1999) 425–441 429
(statically condensed) stiffness matrices, respectively, tal form. For each time step, incremental displacements
and ü, u̇ , and u are the N ⫻ 1 acceleration, velocity, ⌬u are solved and cumulate to obtain the total displace-
and displacement vectors, respectively. The vector F(t) ments.
contains the N ⫻ 1 vector of excitation forces expressed If we express the incremental displacements as
as a function of time. If the excitation is due to ground
motion, F(t) = − müg , where is an N ⫻ 1 unit vector ⌬u = ⌽⌬q (20)
and üg is the ground acceleration.
Although the mass matrix in Eq. (18) is a diagonal where ⌽ is an N ⫻ N modal matrix and ⌬q is an N ⫻
matrix in accordance with the lumped mass model used 1 vector of incremental generalized displacements, Eq.
in the present analysis, both the damping and stiffness (19) can be transformed to a set of N modal equations
matrices include off-diagonal terms. This means Eq. (18) of the form
is a set of N coupled equations for the N-story frame.
Eq. (18) can be solved using various explicit and implicit ⌽Tn
numerical integration schemes [22,23]. Nevertheless, the ⌬q̈n + 2nn⌬q̇n + 2n⌬qn = ⌬Feff(t)
M
method of modal analysis is used here. Modal analysis
allows us to gain better insight into the system response n = 1,2,…,N (21)
by examining the system’s modal responses. Further-
more, modal analysis facilitates the application of codi- where n and n are the damping ratio and natural fre-
fied spectral analysis and design techniques to the prob- quency of mode n, M = ⌽Tm⌽ is the modal mass matrix,
lem at hand. Because modal analysis makes use of the and ⌬Feff(t) = ⌬F(t) + (kinit − kinst)⌬u is the effective
principle of superposition, in a strict sense it is appli- excitation force vector.
cable only to linearly elastic systems. Since the system The above transformation is made possible by the
to be analyzed is nonlinear, a modified form of modal orthogonality property of the modes and by assuming
analysis is employed here. In this modified form, the classical damping for the system. Eq. (21) can now be
effect of nonlinearity is accounted for by subjecting the solved as N single degree of freedom (SDOF) systems,
system to a set of fictitious forces. This method of modal each with a damping ratio of n and a natural frequency
analysis has been successfully applied to solve dynamic of n. For the present study, a fourth order Runge–Kutta
problems involving geometrical nonlinearity [24] and method is used to solve Eq. (21). Once all the modal
material nonlinearity [25,26]. The method is now equations are solved, the incremental displacements can
extended to cover the effect of connection flexibility. be calculated from Eq. (20) and the total displacements
Because of the presence of nonlinearities and connec- obtained from
tion flexibility, the stiffness matrix k in Eq. (18) is not
a constant. To incorporate this change of frame stiffness u = Σ⌬u (22)
into the analysis, Eq. (18) is rewritten as
where kinit = initial stiffness formulated using the initial A Mathcad [16] computer program has been
stiffness properties of the frame, and kinst = instan- developed to perform the above analysis. To verify the
taneous stiffness incorporating any nonlinear and con- program, two example flexibly connected frames will be
nection flexibility effects. In Eq. (19), the last term (kinit analyzed and the results compared with those in the pub-
− kinst)u represents the fictitious force vector. This vector lished literature. The first example is a portal frame
vanishes if the frame remains linearly elastic resulting shown in Fig. 5. The frame was used by Lui and Lopes
in no change in frame stiffness, i.e. if kinit = kinst. The [10] to study the dynamic and seismic responses of
advantage of using a fictitious force vector to handle frames with flexible connections. In Fig. 6 the displace-
nonlinear effects is apparent when one applies the usual ment response curves (Rd = umax/ustatic versus forcing fre-
modal transformation technique to Eq. (19). Because m, quency ) generated using the above fictitious force pro-
c and kinit on the left side of the equation are all con- cedure are compared to those reported in Ref. [10]
stants, the analysis can be carried out by using the initial generated using a step-by-step nonlinear direct inte-
dynamic properties of the frame. gration scheme. The forcing function used was F(t) =
It should be noted that the fictitious force vector in (0.2 mg)sint and the damping ratio used was = 0.05.
Eq. (19) is not known a priori because u is an unknown Three different connection types with connection para-
and kinst is a function of u. One approach is to formulate meters given in Table 1 were used for the comparison.
the fictitious force vector based on information obtained As can be seen, the results compared well with each
from a previous cycle of analysis, and iterate until con- other.
vergence. Also, Eq. (19) is often solved in an incremen- The second example is a three-story frame shown in
430 J.C. Awkar, E.M. Lui / Engineering Structures 21 (1999) 425–441
Table 1
Connection parameters
can be designed more economically than a system with tations, the potential for a more economical design
large base shear. exists.
Figs. 16 and 17 show the base shear time–history
response curves for the five-story frames subjected to the 5.4. Modal responses and modal contributions
San Fernando and the El Centro earthquakes, respect-
ively. It is clear from the figures that there is a reduction One objective of the present study is to examine the
in the maximum base shear for the flexibly connected seismic responses of flexibly connected frames through
frames. Similar results are observed for the eight-story their modal responses. By examining the modal
frames [27]. It is important to note that base shear is not responses, one can gain important insight into the
always proportional to structural stiffness. Rather, it is response characteristics of the frames. Furthermore,
frequency dependent [5,10]. As a rough estimate, if the modal analysis is an important basis for applying many
frequency of the excitation force is higher than the aver- codified spectral analysis and design techniques to seis-
age natural frequencies of two given systems, the mic problems.
maximum base shear will be higher for the system with Figs. 18–21 show the time–history modal responses
the higher stiffness [10]. The opposite is true if the forc- for the roof displacement, top story shear, base shear,
ing frequency is lower than the average natural fre- and base overturning moment for the eight-story semiri-
quencies of the two systems. Because the excitation fre- gid frame subjected to the San Fernando quake. From
quencies of the earthquakes are quite high compared to these figures it can be seen that depending on the
the natural frequencies of the frames, a reduction in base response function, the first mode is not necessarily the
shear occurs for the more flexible frames. Since base predominant mode. Also, to obtain reasonably good
shears (hence the design forces and moments) are lower results for shear and moment, often more than two
for the flexibly connected frames under earthquake exci- modes must be used. These observations are not earth-
J.C. Awkar, E.M. Lui / Engineering Structures 21 (1999) 425–441 433
quake specific. Similar observations are noted if the El quake dependent. For instance, higher mode contri-
Centro quake is used as the ground excitation [27]. butions for the frame with semirigid connections are
To further demonstrate the effect higher modes have more pronounced for the San Fernando quake than for
on the total response of flexibly connected frames, the the El Centro quake. This can be explained by the fact
percent contributions from modes 2–8 for the story that the peak acceleration of the San Fernando quake is
shears of the rigid, semirigid and flexible eight-story much higher than that of El Centro, resulting in the
frames are plotted in Fig. 22. Two observations are ‘yielding’ of more connections in the frame. The semiri-
noted. First, regardless of the earthquake, higher mode gid frame under the San Fernando quake is therefore
contributions become more important as the flexibility more flexible when compared to the frame under the El
of the frames increases. The degree to which connection Centro quake. Second, regardless of connection flexi-
flexibility affects this contribution is, however, earth- bility, higher mode contributions to story shear are more
434 J.C. Awkar, E.M. Lui / Engineering Structures 21 (1999) 425–441
Table 2 important for the lower and especially the upper stories
Natural periods of the five-story frames than for the middle stories. Nevertheless, this phenom-
Natural periods (sec) enon is less noticeable for the frame with flexible con-
nections. For the flexible frame, the contribution from
Mode Rigid Semirigid Flexible the first mode is very small, primarily because the
response is well into the displacement sensitive region
1 1.101 2.216 3.992 of the response spectrum (the region where the response
2 0.329 0.495 0.593
3 0.163 0.202 0.214 diminishes as the frame’s period increases).
4 0.100 0.108 0.110
5 0.067 0.069 0.069 5.5. Use of modal combination for flexibly connected
frames
Fig. 13. Comparison of first six mode shapes for eight-story frames.
436 J.C. Awkar, E.M. Lui / Engineering Structures 21 (1999) 425–441
modal combination for the top story deflection (u8), top (21) are not exactly uncoupled. The coupling occurs in
story shear (V8), base shear (Vb), and base overturning the fictitious force term. Although the errors are within
moment (Mb) of the eight-story frames with the three the expected errors for any modal combination method,
connection types (rigid, semirigid and flexible) are com- care must be exercised in applying the modal combi-
pared to those obtained from the time–history analyses nation method for spectral analysis and design of flexibly
in Tables 4 and 5 for the San Fernando and the El Centro connected frames.
quakes, respectively. Also shown in the tables are the
percent errors incurred and the maximum responses for
each mode. From the tables, it can be seen that while 6. Summary and conclusions
the CQC method produces rather good results for the
rigid frame, relatively large errors can occur for the The seismic responses of flexibly connected multi-
semirigid and the flexible frames. The error seems to story frames were studied using a modified modal analy-
increase with connection flexibility. The larger error for sis procedure wherein the nonlinear effects were incor-
the flexibly connected frames can be attributed to the porated in a fictitious force vector. The numerical
application of a linear combination process to a nonlin- analysis procedure was implemented in a Mathcad pro-
ear problem and that the modal equations shown in Eq. gram and two multistory frames with three connection
J.C. Awkar, E.M. Lui / Engineering Structures 21 (1999) 425–441 437
Fig. 15. Roof displacement responses of five-story frames under San Fernando earthquake.
Fig. 16. Base shear responses for five-story frames under San Fernando earthquake.
438 J.C. Awkar, E.M. Lui / Engineering Structures 21 (1999) 425–441
Fig. 17. Base shear responses for five-story frames under El Centro earthquake.
Fig. 18. Roof displacement modal contributions for eight-story semirigid frame.
J.C. Awkar, E.M. Lui / Engineering Structures 21 (1999) 425–441 439
Table 4
Comparison of peak responses for the eight-story frames under San Fernando earthquake
Mode u8 (m) V8 (105 N) Vb (105 N) Mb u8 (m) V8 (105 N) Vb (105 N) Mb u8 (m) V8 (105 N) Vb (105 N) Mb
(MNm) (MNm) (MNm)
1 0.3281 1.830 8.816 15.33 0.1993 0.216 0.972 1.733 0.3498 0.024 0.093 0.172
2 0.0538 2.730 4.834 0.261 0.0836 1.355 2.547 0.378 0.1548 1.153 2.605 1.243
3 0.0099 1.647 2.280 0.869 0.0254 1.993 2.776 1.246 0.0363 1.958 3.096 1.342
4 0.0018 0.775 1.042 0.144 0.0036 0.928 1.611 0.348 0.0035 0.744 1.539 0.400
5 0.0005 0.424 0.599 0.128 0.0007 0.470 0.918 0.207 0.0007 0.392 0.904 0.199
6 0.0001 0.167 0.425 0.040 0.0002 0.283 0.762 0.120 0.0002 0.192 0.566 0.099
7 ⬇0 0.043 0.224 0.033 ⬇0 0.065 0.356 0.0571 ⬇0 0.043 0.257 0.041
8 ⬇0 0.003 0.213 0.031 ⬇0 0.004 0.247 0.0348 ⬇0 0.004 0.226 0.032
CQC 0.333 3.86 10.5 15.4 0.218 2.70 4.50 2.22 0.385 2.46 4.54 1.91
Actual 0.317 3.91 10.6 15.2 0.227 3.05 4.88 2.37 0.408 2.91 5.41 1.68
% error 5 −1.3 −1 1.3 −4 −11 −7.8 −6.3 −5.6 −15 −16 14
440 J.C. Awkar, E.M. Lui / Engineering Structures 21 (1999) 425–441
Fig. 22. Higher mode responses in story shear for eight-story frames.
J.C. Awkar, E.M. Lui / Engineering Structures 21 (1999) 425–441 441
Table 5
Comparison of peak responses for the eight-story semirigid frames under El Centro earthquake
Mode u8 (m) V8 (105 N) Vb (105 N) Mb u8 (m) V8 (105 N) Vb (105 N) Mb u8 (m) V8 (105 N) Vb (105 N) Mb
(MNm) (MNm) (MNm)
1 0.1149 0.641 3.087 5.368 0.3810 0.413 1.858 3.312 0.2797 0.019 0.074 0.139
2 0.0218 1.105 1.957 0.106 0.0605 0.981 1.844 0.273 0.0610 0.454 1.026 0.490
3 0.0039 0.643 0.889 0.339 0.0093 0.732 1.019 0.458 0.0139 0.750 1.186 0.514
4 0.0008 0.320 0.431 0.060 0.0015 0.373 0.648 0.140 0.0015 0.318 0.658 0.171
5 0.0002 0.181 0.256 0.055 0.0004 0.228 0.445 0.100 0.0003 0.193 0.444 0.098
6 ⬇0 0.0404 0.103 0.010 0.0001 0.179 0.484 0.076 0.0001 0.068 0.201 0.035
7 ⬇0 0.0106 0.068 0.009 ⬇0 0.030 0.168 0.027 ⬇0 0.013 0.096 0.014
8 ⬇0 0.0008 0.052 0.008 ⬇0 0.003 0.167 0.024 ⬇0 0.002 0.080 0.013
CQC 0.117 1.51 3.83 5.38 0.386 1.40 3.00 3.36 0.287 0.968 1.80 0.758
Actual 0.121 1.46 3.65 5.36 0.363 1.60 2.65 3.35 0.265 1.26 1.81 0.626
% error −3.3 3.4 5 0.4 6.3 −13 13 0.3 8.3 −23 −0.6 21