Vous êtes sur la page 1sur 24

EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS, VOL.

22,509-532 (1993)
MODELLING OF MATERIAL NON-LINEARITIES IN STEEL
STRUCTURES SUBJ ECTED TO TRANSIENT DYNAMIC LOADING
A. S. ELNASHAI AND B. A. IZZUDDIN
Civil Engineering Department. Imperial College of Science, Technology & Medicine. Imperial Colfege Road, London S W7 2BU, U. K.
SUMMARY
This paper is concerned with the modelling of the behaviour of steel under cyclic and dynamic loading conditions.
After a general discussion regarding the requirements for accurate and efficient modelling in dynamics, two
models are described and implemented. The bilinear stress-strain constitutive relationship with kinematic hardening is
widely used in many computer codes, hence is used for control purposes. The multisurface plasticity model is said to
exhibit the important qualities of strain hardening, softening and relaxation to a mean stress. This model is described in
detail and notes on model parameter evaluation are given. A number of validation examples are presented, due to the
complexity of implementation of the multisurface formulation. This is followed by comparisons between the bilinear
response predictions and those of the multisurface model for cyclic and dynamic tests on beam-columns. It is concluded
that in the absence of material test data under cyclic loading, the bilinear model provides acceptably accurate response
predictions. However, the multisurface model provides a significantly closer fit to experimental results, due to its ability
to model a yield plateau and a non-linear strain-hardening regime as well as cyclic degradation. It can also beused for
new types of steel where no distinct yield point is observed.
1. INTRODUCTION
Advanced analysis techniques are being increasingly used, and demanded, by engineers facing the task of
understanding complex structural behaviour under severe dynamic loading. This is particularly true for
structures subjected to earthquake loading, where the concept of capacity design requires strict failure mode
control. Moreover, the accurate assessment of section and member ductilities requires a clear definition of the
point of collapse. It is therefore essential to ascertain that computer programs identify structural and not
numerical collapse under complex loading scenarios using material models that reflect the most salient
response characteristics.
Whilst there is a proliferation of software with non-linear dynamic analysis capabilities, researchers and
practitioners who have attempted to conduct collapse analysis recognize that this problem is far from being
solved. Successful analysis up to genuine collapse imposes very stringent conditions on the material,
member and structural levels as well as on the solver and iterative procedures employed.
The multisurface material plasticity characterization used reflects the observed cyclic hardening at high
strain amplitude, softening at low strain amplitude and mean-stress relaxation under a constant strain
amplitude. The verification examples presented not only confirm the accuracy of the model implementation
and formulation but also vindicate the solution procedures used in the context of analysis practicality and
economy. Differences in response due to the use of either the multisurface model or the conventional bilinear
model are discussed and the relative computational efficiency of the two models is highlighted.
0098-8847/93/060509-24$17~00
0 1993 by J ohn Wiley & Sons, Ltd.
Received 14 July 1992
Revised 23 November 1992
510 A. S. ELNASHAI AND B. A. IZZUDDIN
2. MATERIAL RESPONSE OBSERVATIONS
The uniaxial behaviour of mild steel in its virgin state is essentially linear elastic up to the point of yield.
Under monotonic loading, the characteristic yield plateau is followed by a region of increased strength due to
strain hardening, as shown in Figure 1. This behaviour continues, though with a decreasing slope, up to the
ultimate strength, where the tangent to the stress-strain curve tends to the horizontal. Typical values of
strain-hardening strain is about 2&30&,, whilst the ultimate strength is attained corresponding to a strain of
about 2W250cy.
When subjected to constant strain amplitude cycling, mild steel exhibits a response which converges to
a stabilized saturation loop dependent only on the amplitude of cycling, as confirmed by the experiments of
Feltner and Laird' carried out on high stacking fault energy metals, to which structural steel belongs. The
curve joining the tips of the stabilized cycles is known as the cyclic stress-strain curve, as shown in Figure 2.
Due to the difference in shape between the cyclic and virgin curves, the transient response under constant
strain amplitude cycling is characterized by softening for small amplitudes and hardening for large am-
plitudes. Relaxation to zero mean stress accompanies the process of cyclic softening or hardening if, during
the transient response, the mean stress attains a non-zero value, as demonstrated in Figure 3. Therefore,
a reasonably accurate cyclic model for steel must represent the virgin response, the steady-state cyclic
response and the transient behaviour involving softening, hardening, and mean-stress relaxation. Such
a model must also be capable of representing random cycling without spurious prediction of the real
response.
3. REVIEW OF EXISTING MODELS
Several models for structural steel were developed to represent cyclic behaviour under generalized loading in
the multiaxial stress-strain space. In the current study, such models are considered only in the uniaxial
context, and comparisons are made with other models exclusively developed for the uniaxial case.
stress t
Figure 1. Virgin monotonic response of mild steel
Figure 2. Steady-state cyclic response of mild steel
TRANSIENT DYNAMIC LOADING 511
t
Stress
Yiel
Stabilized cycle
d%
Iv
II
b
I I! Strain
Softenine under small strain amplit&
t
Stress
Hardenine under 1 - in amolitudps
Mean-stress at
I Stabilized cycle
Mean-stress relaxation
Figure 3. Transient response of mild steel
Dafalias and Popovz discussed the relative advantages of kinematic and isotropic hardening models. The
authors pointed out that purely isotropic hardening is not suitable for modelling stress reversals with
Bauschinger effect3 since such a hardening rule is not associated with any direction of loading. They also
presented a multiaxial model employing the concept of a bounding surface enclosing the yield surface, and
assumed that the proximity of the two surfaces in the course of their coupled translation and deformation
during plastic loading determines the value of the plastic modulus. A purely kinematic case of the model was
suggested for uniaxial loading, and a procedure for the calculation of the model parameters from experi-
mental results was demonstrated. Despite the computational advantage of such a model, it does not
represent accurately the virgin response and it predicts a sudden reduction in the elastic stress range upon
detection of plasticity. The latter disadvantage is mainly due to neglecting isotropic hardening, hence the
independence of the Bauschinger effect of the stress-strain history.
512 A. S. ELNASHAI AND B. A. IZZUDDIN
Petersson and Popov4* presented an extension and a refinement of the Dafalias-Popov model, motivated
by the spurious prediction that the model exhibits if unloading and reloading are involved without
substantial plastic flow in the opposite sense. The authors pointed out that this deficiency can be remedied by
the use of a number of intermediate surfaces between the yield and bounding surfaces, and presented a new
model based on this multisurface approach. The multisurface model is capable of representing the virgin
response of steel. Through the use of a weighting function applied to the virgin and cyclic stress-strain curves,
it also takes account of the gradual transition from virgin to cyclic behaviour as a function of accumulated
plasticity. The ability of the multisurface model to represent cyclic softening, cyclic hardening and mean-
stress relaxation is demonstrated in subsequent sections of this paper, as discussed in more detail by
Izzuddin.6
Santhanam' presented a uniaxial piecewise linear model employing two parameters for cyclic stiffness
degradation and yield stress growth. Despite its simplicity, this model has shortcomings in that it is unable to
represent the basic cyclic phenomena of softening and mean-stress relaxation. Also, it has the inherent
disadvantage of predicting spurious behaviour if unloading and reloading are involved without substantial
plastic flow in the opposite sense, similar to the model of Dafalias and Popov.
A two-surface definition similar to that used in the Dafalias-Popov model was employed by Tseng and
Lee,* but they assumed only isotropic hardening for the bounding surface which expands upon contact with
the enclosed yield surface. The authors suggested that, for structural steel, the hardening of the bounding
surface is obtained from the virgin response, while the size of the yield surface is allowed to diminish with the
accumulation of plasticity to reflect the dependence of the Bauschinger effect on the stress-strain history. The
authors also demonstrated the ability of the proposed model to represent the phenomena of cyclic softening,
cyclic hardening and mean-stress relaxation. However, an implication of the use of isotropic hardening for
the bounding surface is the dependence of the stabilized constant strain amplitude cycles on the stress-strain
history, which contradicts experimental findings. Also, the proposed model can predict spurious results if
unloading and reloading are involved without substantial plastic flow in the opposite sense.
A uniaxial cyclic model was proposed by Cofie and Krawinkler.' This made use of the Dafalias-Popov
relationship for the determination of the plastic modulus. The authors suggested that the bounding lines are
updated upon stress reversal so that cyclic softening, cyclic hardening and mean-stress relaxation can be
accounted for. While the experimental results indicated that softening, hardening and mem-stress relaxation
factors depend on the number of excursions already performed and the cyclic strain amplitude, the authors
adopted constant values for these factors to avoid modelling complexities. In a discussion, White and
McGuire" suggested that improvements to the model are needed so that Bauschinger-type effects become
dependent on the stress-strain history, and a variable factor in the Dafalias-Popov equation is used such that
small amplitude cycles approach the bounding lines more abruptly.
Castiglioni' compared the Ramberg-Osgood model modified by Ma et al." with the Menegotto-Pinto
model modified by Filippou et The author pointed out the computational advantage of the
Menegotto-Pinto model, in that it does not require an iterative procedure to determine the stress corres-
ponding to a given strain. While both models predict spurious behaviour in the case of partial unloading, the
use of the Masing hypothesis, with the Ramberg-Osgood model, may lead to a significant overestimate of the
response. Castiglioni concluded from comparisons with the experimental results that the model of Ma et a/.
gives better correlation with the virgin response, while the model of Filippou et al. is more suitable for cyclic
loading. However, although the modification of Filippou et al. of the Menegotto-Pinto model involved the
addition of an isotropic hardening parameter, such a parameter does not relate directly to the gradual
softening or hardening behaviour observed under constant strain amplitude cyclic loading.
In this paper, the bilinear model with kinematic hardening and the Petersson-Popov multisurface model
are implemented, with modifications, and linked to the space frame analysis program ADAPTIC described
by Izzuddin6 and Izzuddin and E1na~hai .l ~ A cubic elastoplastic finite element is used to investigate the effect
of material response characterization on the results obtained from collapse analysis of steel and composite
members. Since most existing analysis environments utilize the bilinear model with kinematic strain
hardening to represent the stress-strain curve of mild steel, the equations and results of this model are given
for comparison and to highlight the difference in structural response parameters due to differences in
stress-strain curve description.
TRANSIENT DYNAMIC LOADING 513
4. BILINEAR KINEMATIC MODEL
In this model, the elastic range remains constant throughout the various loading stages, and the kinematic
hardening rule for the yield surface is assumed to bea linear function of the increment of plastic strain, as
shown in Figure 4.
Mathematically, the calculation of the current stress state is expressed as follows:
u =uo
Q =ao +E( E - EO)
} if( u0 - ay - a ' ) < ( E - - ~ o ) < ( uo +ay - ao )
where the symbols are defined according to Figure 5.
"t
Figure 4. Loading and unloading paths of bilinear kinematic model
I
Positive strain ingernent Neeative strain incremenf
Figure 5. Stress determination with the bilinear kinematic model
514 A. S. ELNASHAI AND B. A. IZZUDDIN
The tangent modulus, required for updating the element stiffness matrix, is dependent on the current stress
state, and is given by
E if (a - oy) <o -=(01 +by)
pLE if o =( a - oy) or o =(a +oy)
5. MULTISURFACE MODEL
5. 1. Constitutive relationships
The uniaxial case of the multisurface model proposed by Petersson and Popov4q5 is employed to represent
the cyclic response phenomena of mild steel. A number of surfaces enclosing the yield surface are each
associated with a value for the increment of plastic strain, and the expansion/contraction (isotropic
hardening) and translation (kinematic hardening) of these surfaces are governed by pre-specified hardening
rules.
Figure 6(a). Configuration of loading surfaces before increment of plastic strain
Figure 6(b). Configuration of loading surfaces after plastic strain increment
E P3 E p z ~ / ~
Figure 6(b). Configuration of loading surfaces after plastic strain increment
TRANSIENT DYNAMIC LOADING 515
At the start of a strain increment, the configuration of the loading surfaces may be as shown in Figure 6(a):
t i p =K O ( E , ~ ) =initial radius of loading surface ( i )
a? =ao( cpi ) =initial centre of loading surface ( i )
cpi =increment of plastic strain
(3)
If the application of a strain increment results in plastic strain, the loading surfaces expand/contract and
translate in the direction of loading, as shown in Figure 6(b), according to
E .
P
E
Pi
"t
%
Figure 7. The basic curves for isotropic hardening determination
516 A. S. ELNASHAI AND B. A. IZZUDDIN
where, the (+) depends on the sign of the plastic strain increment Acp, and
K~ =K ( E ~ ~ ) =current radius of loading surface ( i )
ai =a(spi) =current centre of loading surface ( i )
Asp =current increment of plastic strain
( 5)
Peterson and Popov suggested that isotropic hardening of the loading surfaces is expressed as a function
of accumulated plasticity. A weighting function, dependent on the cumulative plastic strain, is hence applied
to the virgin and cyclic stress-strain curves to obtain the initial and current sizes of the loading surfaces
(Figure 7):
Ko(&pi) =Kb(&pi) +w ( E i ) x LKa(&pi) - Kb( &pi ) l
[%(&pi) - Kb(&pi)]
(6)
K(&pi) =Kb(&pi) +w(Ep)
1
Int. 1 Int. 2 Int. 3 Int.4 Int.5 Int.6 Zp
Figure 8. Polynomial representation of the basic curves of the multisurface model
TRANSIENT DYNAMIC LOADING 517
in which
K , ( E ~ ~ ) =virgin stress-strain curve
K ~ ( E ~ ~ ) =cyclic stress-strain curve
W(Ep) =weighting function
to
E; =1 J A E ~ J =cumulative plastic strain at start of increment
Ep =1 IAcPl =cumulative plastic strain at end of increment
0
f
0
(7)
5. 2. Functional representation
In the process of updating the centres of the loading surfaces according to equation (l), a continuous
representation of the surface sizes ( K O ) as a function of cpi is required, so that any increment of plastic strain
( AE ~ ) can be allowed. From equation (6), this can be guaranteed if a continuous functional representation is
adopted for the virgin and cyclic stress-strain curves, as well as the weighting function.
In this paper, each of these curves is represented by five cubic polynomials over the adjacent intervals and
a straight line at the end, as shown in Figure 8. Conditions of continuity of values and slopes are enforced at
the common points to obtain the parameters of the cubic polynomials and the straight line.
5.3. Determination of stresses
For an increment of strain (AE), the post-yield loading curve to be considered is given by
bp(Epi) =ao( Epi ) f Ko(Epi) (8)
The linear elastic stress-strain relationship is employed if the current stress state remains within the yield
where the f depends on the sign of At.
surface:
B =(TO +E AE if 1 0 1 6 lop(0)l
(9)
However, if condition (9) is not satisfied, plastic strain is induced, and the post-yield curve of equation (8)
must be employed. Hereafter, the stress state is assumed to lie initially on the yield surface in order to simplify
the following equations. When the stress state is initially within the yield surface, the part of A& required to
bring the stress state to the yield surface is first obtained, before the remainder of the increment As is applied.
The total strain increment consists of elastic and plastic increments; hence the following relationships hold:
AE =A&, +Atp
B =uJO) +E BE,
(10)
B =B,(AE,)
This represents a system of three equations with three unknowns: B, AE, and A E ~ . However, since gP can be
a cubic function of Aep, or cpi, according to the previous section, an iterative procedure is required to obtain
the exact solution of equation (lo), which may lead to an excessive computational effort. In this work, a cubic
polynomial fit of B as a function of AE is employed in order to avoid iterations, as shown in Figure 9, and the
parameters of this polynomial can be obtained from its values and slopes at the interval extremities, given by
the following equations:
518 A. S. ELNASHAI AND B. A. IZZUDDIN
O t Cubic fi t (exaggerated) -$?( AE 2)
. .
Post-vield curve o f stress versus total strain in-
Figure 9. Determination of stress from total strain increment
Since the actual virgin and cyclic stress-strain curves are modelled using a number of polynomial functions
according to the previous section, the exact solution of equation (10) does not guarantee exact correlation
with the actual response. For this reason, no effort is dedicated to obtaining the exact solution of
equation (lo), and the fitting procedure employing a cubic polynomial with parameters as given by equation
(1 1) is observed to yield reasonably accurate results while avoiding excessive iterative computations.
5.4. Tangent modulus
The tangent modulus is required for the updating of the element stiffness matrix. If the current stress state
is within the yield surface, the linear elastic modulus is used. Otherwise, the tangent modulus is determined
TRANSIENT DYNAMIC LOADING 519
from the elastic and plastic moduli. This is established by considering that the infinitesimal increment of total
strain is composed of elastic and plastic parts resulting in the same infinitesimal increment of stress when
multiplied by their respective moduli. This results in the following expression for the tangent modulus:
E if I4 -=lqJ (0)I
- +1/(dq&J (A&p) if 142 lap(0)I
1 / E
*.-I d& 1 (12)
where the function op) represents the post-yield loading curve as befined in equation (8).
5.5. Evaluation of model parameters
Whilst capable of representing the detailed behaviour of a class of steel, by virtue of the presence of a yield
point, plateau and non-linear hardening region, as well as cyclic response characteristics, the multisurface
plasticity model requires a relatively elaborate calibration procedure. The procedure for calibration, using
experimental measurements, is as follows:
(i) The monotonic curve. This may be obtained from a standard tensile test giving the yield point, the yield
plateau and the non-linear hardening regime. From this curve, the parameters K, i (Figure 8) are evaluated.
(ii) The cyclic curve. Testing at various constant strain amplitudes is needed. At each strain amplitude,
cycling should continue until a stabilized response is obtained. A plot of half the cycle size versus the plastic
strain would yield the parameters KbO, Kbl , Kb2, etc. (Figure 8) required for model definition.
(iii) The weightingfunction. Cyclic testing at low, but plastic, strain amplitude is necessary here. For each
loading excursion, the point at which yielding starts should be monitored. I n which case, equations similar to
equation (6) may be used, where the only unknown is the weighting function W (Figure 8), since the
K parameters have already been evaluated as above.
In the absence of adequate experimental data, or the facility to conduct the necessary testing, a process of
calibration may be used on a structural level. This comprises the choice of model parameters to improve the
comparison between the load-displacement response of a single structural member, then testing the model
predictions in comparison to other members, with no further changes in the parameters. This is indeed the
process followed for the application to cyclic and dynamic testing of beam columns presented in Section 6.
5.6. Verification
Properties of the multisurface model. The predictions of the multisurface model under constant strain
amplitude cycling are studied for a material with virgin and cyclic stress-strain responses as in Figures 10(a)
and 10(b). The transition between the virgin and cyclic behaviour is governed by the weighting function
dependent on the accumulated plastic strain, as shown in Figure IO(c).
Under small amplitudes of cyclic strain, the multisurface model is capable of representing the softening
behaviour by virtue of the accumulated plastic strain and its effect on the weighting function, as demon-
strated in Figure 10(d). For large amplitudes of cyclic strain, the multisurface model can predict the
hardening behaviour, again through the use of the weighting function, as shown in Figure 10(e).
The model also caters for the process of mean-stress relaxation. However, since this process is associated
more with the translation of the loading surfaces (i.e. kinematic hardening), the mean stress does not relax
exactly to a zero value but rather to a line of slope equal to the hardening modulus of the bounding surface, as
observed in Figure 10(f).
Tests by Mizuno et al. The cyclic axial response of the tubular beam column depicted in Figure 11 (a) is
obtained using the multisurface model. For the purpose of comparison with the results of Mizuno et al., the
cyclic stress-strain curve (Kb) is employed from the start of analysis. The loaddisplacement curves plotted in
Figure 1 1 (b) demonstrate reasonable agreement with the prediction of Mizuno et al., while the slight
520 A. S. ELNASHAI AND B. A. IZZUDDIN
700
600
n m
1400 .
2
w 3 0 0
m
100
0
0.00 0.01 0.02 0.03 0.04 0.05 0.06
Pi
Y
e
r'igure 10(a). Virgin stress-strain curve
700
600
n 5 0 0
N
. L o o
5 , 3 0 0 a
Y
200
100
0
0.00 0.01 0.02 0.03 0.04 0.05 0.06
Pi
Figure 10(b). Cyclic stress-strain curve
1.2
1 .o
0.8
B 0.6
0.4
0.2
0.0
0.00 0.01 0.02 0.03 0.04 0.05 0.06
-
EP
Figure lO(c). Weighting function
TRANSIENT DYNAMIC LOADING
600-
400-
3 200-
0-
a
E
i!
x -200 -
-400 -
-600
52 1
I I
-0.02 -0.01 0.00 0.01 0.02
Strain
Figure 10(d). Prediction of softening with multisurface model
-0.02 -0.01 0.00 0.01 0.02
Strain
Figure 10(e). Prediction of hardening with multisurface model
( Slope=O.OlE/( 14.01) )
-0.02 -0.01 0.00 0.01 0.02
Strain
Figure 10(f). Prediction of mean-stress relaxation with multisurface model
522 A. S. ELNASHAI AND B. A. IZZUDDIN
(-)
@114x 2 3 m2
400
100
0
O.OO0 0.005 0.010 0.015 0.020 0.025 0.030 0.035 0.040
E pi
Figure I 1 (a). Geometry and material properties of fixed-ended beam column
0
Y 3
100
p-100
- 200
i -
-50 -40 -30 -20 -10 0 10 20 30 40 50
A (mm)
Figure 1 1 (b). Cyclic axial response of fixed-ended beam column
disagreement at the load peaks may be due to differences in the degree of accuracy in representing the cyclic
curve.
6. APPLICATIONS
The models described in Sections 4 and 5 were implemented in the space frame analysis program ADAPTIC,
developed to predict the static and dynamic responses of framed structures by using mesh adaptation,
automatic solution strategy and time-step adjustment. Whereas the use of the plastic hinge formulation
provides the most efficient and a sufficiently accurate solution, a section analysis approach is necessary in
order to investigate the stress-strain relationships presented above. For this purpose, the constitutive
relationships were implemented within a beam-column finite element with a cubic polynomial shape function
TRANSIENT DYNAMIC LOADING 523
and employing a second-order Gaussian quadrature rule. The full derivation of the elastoplastic cubic
element matrices is given l 6 The analyses presented hereafter are intended to demonstrate the
accuracy of the current implementation of the multisurface plasticity model, the differences between use of
the latter model and the conventional bilinear kinematic hardening relationship and the relative computa-
tional effort involved in analysis using both models.
6.1. Sway frame
The hypothetical sway frame shown in Figure 12(a) is assumed to have the material properties used in the
example of Section 5.5. The frame is analysed under constant displacement amplitude cycling, and the
response is obtained using the bilinear model (oy =450 N/mm2, p =0018) and the multisurface model.
The load-displacement curves of Figure 12(b) show that, with the bilinear model, cyclic stabilization is
achieved after the first cycle, while, with the multisurface model, stiffness degradation is observed as plasticity
accumulates, and hardening is predicted in the cyclic stress amplitudes. Although the multisurface model
gives a better representation of the actual material response, a comparison of CPU times demonstrates that
the bilinear model is computationally more efficient (5.5 min for the bilinear model compared to 11.25 min
for the multisurface model; running on a Micro Vax I1 with 8 Mb of memory).
6.2. Cyclic and dynamic response of beam-columns
A number of steel and composite steel/concrete beam columns were tested at Imperial College,17-19 and at
the Institute of Industrial Science, Tokyo.20 The tests were conducted using cyclic loading and pseudo-
dynamic testing under earthquake strong-motion.
E
E
L
3
2
1
0
- 1
-2
-3
Figure 12(a). Geometric configuration of sway frame
-0.15 -0.10 -0.05 0.00 0.05 0.10 0.15
A (m)
Figure 12(b). Cyclic response of sway frame
524 A. S. ELNASHAI AND B. A. IZZUDDIN
Cyclic and pseudo-dynamic tests were conducted on composite steel/concrete members and comparison
with ADAPTIC results was undertaken. The test-rig arrangement used is shown in Figure 13(a), whilst the
dimensions and material properties of the tested member is shown in Figure 13(b). In Figures 14(a) and 14(b),
the cyclic experimental results for force-displacement (Reference 19; model ICA2) are depicted as well as for
the case of the bilinear model. The corresponding multisurface results are given in Figure 14(c). Whereas the
bilinear model predicts the overall response and ultimate load accurately, the roundedness of the experi-
mental results are captured only by the multi-surface model. This may have implications on the transient
response, as discussed below.
Results are also compared for the case of earthquake loading under the widely used NS component of the
Imperial Valley (El Centro, 1940) earthquake. For model ICA3, discussed in Reference 19, the
load-displacement response from the test, the bilinear model and the multisurface model are shown in
Figures 15(a)-(c). Whilst the comparison of capacities is very good, the overall shape of hysteresis loops from
the multisurface model are significantly closer to the experiments. Also, the maximum displacements,
particularly near the end of the time history, are more over-estimated by the bilinear model. The transient
response comparison, shown in Figures 16(aHc), confirms the observations from the loaddisplacement
response; both models give good comparisons, with the multisurface results closer to the experiments.
I t is particularly interesting to note that the accuracy of both models deteriorates near the end of the
analysis. This is attributed to different reasons. In the case of the bilinear model, the degrees of freedom
available for calibration are extremely limited; only the strain-hardening modulus may be altered, depending
on the maximum strain reached in an analysis. For the multisurface model, full calibration, as discussed in
a previous section of this paper, requires a number of cyclic material tests, which were not conducted herein.
In which case, the least accurate would be the cyclic curve; hence, progressively deteriorating performance
would be observed as the number of cycles increases.
Figure 13(a). Test-rig for cyclic and pseudo-dynamic loading (Broderick )
TRANSIENT DYNAMIC LOADING 525
Applied Load
47r
55Ox5Sh35 pl.
J
5 no. 6& dia.
straight links
@ 8Omm cs.
9 straight no. ~ 6mm links dia.
@ 4ommcs.
m
N
2
2
2
N
84
- l h weld
Concrete
Compressive Strength 28 MPa
1
%$289 MPa; Ultimate 459 Mpa
cr oss-bar s
Yield 347 Mpa; Ultimate 454 Mpa
nm
6mm dia. 152x152~23 UC
straight link
Concrete
Spot-weld
A - A
12mm weld
550x250~70 pl. T 6 5 m m
I 1 . 1
Figure 13(b). Beam-column specimens for cyclic and pseudo-dynamic testing (Broderickzl )
In test ICA4-Figures 17 and 18(aHcFhysteresis loops and displacement time-histories are compared
with measurements from the tests conducted by Broderick. Here, the comparison between multisurface
plasticity and bilinear predictions shows that the former is far superior, especially when considering the
time-history comparison. This is due to the maximum displacement amplitude reached in this test, which is
higher than ICA3, and the larger permanent plastic deformation; in ICA3, the displacements between time
6 and 8 s oscillated about a permanent set of 5 mm, while in ICA4 this was approximately 25 mm. The
bilinear model incorrectly predicted that the model has recovered to a near-zero mean, whilst the oscillation
about a permanent plastic displacement was predicted accurately by the multisurface model. Table I sum-
marizes parameters used for earthquake assessment of structural systems for the three comparison cases.
Several important points, worthy of discussion, arise from the aforementioned results. Firstly, it is clear
that the multisurface model exhibits the capability of duplicating the experimental results more accurately,
both quantitatively and qualitatively. Of particular interest in seismic design is the length of the plastic hinge.
526 A. S. ELNASHAI AND B. A. IZZUDDIN
100
3 0
i2
8
-100
-100 0 100
displacement (mm)
Figure 14(a). Test ICA2: Experimental forcedisplacement response
-100-
-100 0 1
displacement (mm)
K)
Figure 14(b). Test ICA2: Analytical (bilinear model) forcedisplacement response
-100.-
-100 0 1
displacement (mm)
0
Figure 14(c). Test ICAZ: Analytical (multisurface model) force-displacement response
As shown in Table I, the bilinear model grossly underestimates the plastic hinge length (defined as the length
over which the yielding criterion is satisfied), in contrast to the good predictions of the multisurface model. In
all results, the multisurface model predictions are between 5 and 47 per cent more accurate than the bilinear
model results. However, the effect of model parameters on the response, particularly the energy absorption
capacity, is significant. It is, therefore, important to calibrate the model parameters to reliable virgin and
TRANSIENT DYNAMIC LOADING 527
100
5 0
(r 1
- 100, . v I . , . . . . I
-60 0 60
displacement (mm)
Figure 15(a). Test ICA3: Experimental force-displacement response
- 1 0 0 - 1
-60 0 60
displacement ( v)
Figure 15(b). Test ICA3: Analytical (bilinear model) force-displacement response
- 1 0 0 - 4
-60 0 60
displacement (mm),
Figure 15(c). Test ICA3: Analytical (multisurface model) force-displacement response
cyclic material curves. Notwithstanding, it is recommended to use multisurface formulation where possible,
especially for large strain amplitude variable cycling, since the bilinear model may result in unrealistic
estimates of the ductility, collapse and energy absorption capacity. For small amplitude transient loading,
results obtained from bilinear models are adequately representative of the true behaviour.
528 A. S. ELNASHAI AND B. A. IZZUDDIN
time (seconds)
Figure 16(a). Test ICA3: Experimental time-displacement response
Figure 16(b). Test ICA3: Experimental and analytical (bilinear model) time-displacement responses
ti mc (scconds)
Figure 16(c). Test ICA3: Experimental and analytical (multisurface model) timeedisplacement responses
7. CONCLUSIONS
Whereas modelling of steel plasticity using a bilinear kinematic hardening model is adequate for many
applications in static analysis, from a conceptual point of view, the bilinear model has three main shortcom-
ings in its inability to model: (i) the presence of a horizontal yield plateau followed by a strain-hardening
zone, (ii) the reduction in strain-hardening slope with the increase in strain amplitude and (iii) experimentally
TRANSIENT DYNAMIC LOADING 529
-80 0 80
displacement (mm)
Figure 17(a). Test ICA4 Experimental force-displacement response
Figure 17(b). Test ICA4: Analytical (bilinear model) force-displacement response
-80 0
displacement (mm)
Figure 17(c). Test ICA4 Analytical (multisurface model) force-displacement response
observed cyclic degradation. Under such conditions, the model proposed by Popov and Petersson, the
multisurface plasticity formulation, provides an appropriate modelling tool for use in structural analysis
software packages, especially for earthquake applications. The model is capable of accounting for the effect of
small and large amplitude cyclic loading on softening and hardening, respectively, as well as for the
mean-stress relaxation. It is sufficiently economical, provided that internal iterations to evaluate the stress
530
B .
-20
A. S. ELNASHAI AND B. A. IZZUDDIN
Id
- 4 0 , . . . ,
0 1 2 3 4 5 6 1 8
time (seconds)
, . , , , . . , , , . , , , . . , , , . . , , , , , , , .
Figure 18(b). Test ICA4: Experimental and analytical (bilinear model) time-displacement responses
80
60
40
v
-
g 20
- P
D O
B
-20
-40
0 1 2 3 4 5 6 7 8
time (seconds)
Figure 18(c). Test ICA4: Experimental and analytical (multisurface model) time-displacernent responses
vector can be avoided, as discussed above. Typically, the multisurface solution requires approximately
double the computing time of the bilinear solution. However, both solutions are extremely efficient when
employing automatic mesh adaptation, automatic solution strategy management and automatic time-step
adjustment, as in the Imperial College computer code ADAPTIC.
TRANSIENT DYNAMIC LOADING 53 1
Table I. Comaprison of eathquake response parameters
Test I CA2 I CA3 ICA4
Model Bilinear Multisurface Bilinear Multisurface Bilinear Multisurface
PX 1.07 1.01 1.58 1.35 1.29 1.16
1.06 1 .oo 1.57 1.35 1.35 1.17
0.57 1.04 0.75 1.20 0.5 1 0.82
Values are given as ratios of analytical predictions to experimental measurements.
Comparisons between analytical results using the multisurface plasticity and the kinematic hardening
bilinear models indicate a significant difference in response parameters. In comparison with test data, it is
observed that the multisurface model is superior to the bilinear model in duplicating the experimental data.
However, the accuracy of the former model is dependent on the rigour of the process of calibration to the
measured virgin and cyclic curves, as well as the accuracy of the weighting function.
To conclude, in the context of capacity design under earthquake loading, it is important to pin-point the
onset of plastic flow, plastic redistribution and reduction in load-bearing capacity in order to evaluate, with
a high degree of exactitude, the section, member and system ductilities. It is therefore important to utilize
stress-strain models that exhibit the important behavioural patterns observed in monotonic, variable
amplitude cyclic and dynamic testing. Such phenomena are well-catered for by the multisurface plasticity
model implemented and verified in this paper.
E
P
P i
P$
1:
UY
E
0
a
fca
W
superscript 0
superscript e
superscript p
fcb
NOTATION
Youngs modulus
strain hardening parameter
ratio of analytical to experimental displacement ductility
ratio of analytical to experimental rotation ductility
ratio of analytical to experimental plastic hinge length
initial yield stress (radius of yield surface)
current strain
current stress
current centre of elastic range (centre of yield surface)
virgin stress-strain curve
cyclic stress-strain curve
weighting function
denotes values at the start of an increment
denotes elastic component
denotes plastic component
REFERENCES
1. C. E. Feltner and C. Laird, Cyclic stress-strain response of f.c.c. metals and alloys, Acta metallurgica 15, 1621-1653 (1967).
2. Y. F. Dafalias and E. P. Popov, Plastic internal variables formalism of cyclic plasticity, J. appl. mech. ASME 43, 64-651 (1976).
3. S. P. Timoshenko, Strength of materials, Part 11, 3rd edn, pp. 41 1416, Van Nostrand, Princeton, NJ , 1956.
4. H. Petersson and E. P. Popov, Constitutive relations for generalized loadings, J . eng. meck. di u. ASCE 103, 611-627 (1977).
5. E. P. Popov and H. Petersson, Cyclic metal plasticity: experiments and theory, J . eng. mech. diu. ASCE 104, 1371-1388 (1978).
6. B. A. Izzuddin, Nonlinear dynamic analysis of frames structures, Ph.D. Thesis, Imperial College, University of London, London,
7. T. K. Santhanam, Model for mild steel in inelastic frame analysis, J . struct. diu. ASCE 105, 199-220 (1979).
8. N. T. Tseng and G. C. Lee, Simple plasticity model of two-surface type, J. eng. mech. ASCE 109, 795-810 (1983).
9. N. G. Cofie and H. Krawinkler, Uniaxial cyclic stress-strain behavior of structural steel, J. eng. mech. ASCE 111, 1105-1 120 (1985).
10. D. W. White and W. McGuire, Discussion on: Uniaxial cyclic stress-strain behavior of structural steel, by Cofie 8c Krawinkler
1991.
1985, J . eng. mech. ASCE 113, 1803-1807 (1987).
532 A. S. ELNASHAI AND B. A. IZZUDDIN
11. C. A. Castiglioni, Numerical simulation of steel shapes under cyclic bending: effect of the constitutive law of the material,
Costruzioni metalliche 154-175 (1987).
12. S. Y. M. Ma, V. V. Bertero and E. P. Popov, Experimental and analytical studies on the behavior of reinforced concrete rectangular
and T-beams, Report No. UCE/EERC-76/2, University of California, Berkeley, CA, 1976.
13. F. C. Filippou, E. P. Bertero and E. P. Popov, Effects of bound deterioration on hysteretic behavior of reinforced concrete joints,
Report No. UCB/EERC-83/19, University of California, Berkeley, CA, 1983.
14. B. A. Izzuddin and A. S. Elnashai, ADAPTIC: a program for the adaptive dynamic analysis ofspace frames, Report No. ESEE-89/7,
Imperial College, London, 1989.
15. E. Mizuno, M. Kato and Y. Fukumoto, Multi-surface model application to beam-columns subjected to cyclic loads, J . construct.
steel res. I, 253-277 (1987).
16. B. A. Izzuddin and A. S. Elnashai, 1992, Elastoplastic analysis of steel frames: Part I, Plastic hinge approach; Part 11, Distributed
plasticity approach, ICE, (submitted).
17. A. Y. Elghazouli, Earthquake resistance of composite beam-columns, Ph.D. Thesis, Imperial College, University of London,
London, 1991.
18. A. Y. Elghazouli, A. S. Elnashai and P. J . Dowling, Experimental behaviour of ductile partially encased composite beam-columns,
Earthquake, Blast and Impact; Measurements and Efects of Vibrations Conference, SECED, Manchester, September, pp. 21 1-220,
1991.
19. B. Broderick and A. S. Elnashai, Testing and analysis of partially-encased beam-columns under combined earthquake and axial
loading, Report No. ESEE-92/6, J uly, Imperial College, London, 1992.
20. A. S. Elnashai, K. Takanashi, A. Y. Elghazouli and P. J . Dowling, Experimental behaviour of partially encased composite
beam-columns under cyclic and dynamic loads, Proc. instn civil engrs, London, Part 2 91, 259-272 (1991).
21. B. Broderick, Seismic testing, analysis and design of composite frames, Ph.D. Thesis, Imperial College, University of London (to be
submitted), 1993.

Vous aimerez peut-être aussi