Академический Документы
Профессиональный Документы
Культура Документы
NIGEL PRIESTLEY
Centre of Research and Graduate Studies in Earthquake Engineering and Engineering Seismology (Rose School),
Istituto Universitario di Studi Superiori (IUSS) Pavia, Italy
MICHELE CALVI
LORENZA PETRINI
CLAUDIO MAGGI
Abstract:
Key words: damping, inelastic response, time-history analysis, dynamic test, bridge pier
1. INTRODUCTION
It is common to specify a level of elastic damping in inelastic time-history analysis
(ITHA) to represent damping in the initial stages of response, before hysteretic damping
is activated. This is normally specified as a percentage, typically 5%, of critical damping.
There are a number of ways this damping could be defined, but the principal difference is
whether the damping force is related to the initial or tangent stiffness.
Typically, research papers reporting results on single-degree-of-freedom (SDOF)
ITHA state that 5% elastic damping was used, without clarifying whether this has been
related to the initial or tangent stiffness. With multi-degree-of-freedom (MDOF)
analyses, the situation is often further confused by the adoption of Rayleigh damping,
which is a combination of mass-proportional and stiffness-proportional damping. It is
our understanding that many analysts consider the choice of the initial elastic damping
model to be rather insignificant for either SDOF or MDOF inelastic analyses, as the
effects are expected to be masked by the much greater energy dissipation associated with
hysteretic response. This is despite evidence by others (e.g. Otani (1981)) that the choice
of initial damping model between a constant damping matrix and tangent-stiffness
proportional damping matrix could be significant, particularly for short-period structures.
With initial stiffness elastic damping, the damping coefficient is constant throughout the
analysis, even in the inelastic range of response, and is based on the initial elastic stiffness.
With tangent-stiffness damping the damping coefficient is proportional to the
instantaneous value of the stiffness and it is updated whenever the stiffness changes.
With the case of elasto-plastic response, the tangent-stiffness damping force will be zero
while the structure deforms along a yield plateau.
There are three main reasons for incorporating elastic damping in ITHA:
The assumption of linear elastic response at force-levels less than yield: Many
hysteretic rules make this assumption, and therefore do not represent the
nonlinearity, and hence hysteretic damping within the elastic range for concrete
and masonry structures, unless additional elastic damping is provided.
Foundation damping: Soil flexibility, nonlinearity and radiation damping are not
normally incorporated in structural time-history analyses, and may provide
additional damping to the structural response.
Non-structural damping: Hysteretic response of non-structural elements, and
relative movement between structural and non-structural elements may result in
an effective additional damping force.
Discussing these reasons in turn, it is noted that hysteretic rules are generally calibrated to
experimental structural data in the inelastic phase of response. Therefore additional
elastic damping should not be used in the post-yield state to represent structural response
except when the structure is unloading and reloading elastically. If the hysteretic rule
models the elastic range nonlinearly (as is the case for fibre-element modelling) then no
additional damping should be used in ITHA for structural representation. It is thus clear
that the elastic damping of hysteretic models which have a linear representation of the
elastic range, and which hence do not dissipate energy by hysteretic action at low force
levels would be best modelled with tangent-stiffness proportional damping, since the
elastic damping force will greatly reduce when the stiffness drops to the post-yield level.
It should, however, be noted that when the post-yield stiffness is significant, the elastic
damping will still be overestimated. This is particularly important for hysteretic rules such
as the modified Takeda degrading stiffness rule which has comparatively high stiffness in
post-yield cycles.
If the structure deforms with perfect plasticity, then foundation forces will remain
constant in the structural post-yield stage, and foundation damping will cease. It is thus
clear that the effects of foundation damping in SDOF analysis are best represented by
tangent stiffness related to the structural response, unless the foundation response is
separately modelled by springs and dashpots.
It is shown elsewhere (Priestley et al (2007)) that non-structural damping in modern
buildings is likely to contribute no more than 1% effective damping in the inelastic range
of response. With bridges it is hard to see where non-structural damping could originate.
It is instructive in determining the influence of alternative elastic damping models to
consider the steady-state, harmonic response of an inelastic SDOF oscillator subjected to
constant sinusoidal excitation. This enables direct comparison between hysteretic and
elastic damping energy, and also between elastic damping energy using a constant
damping coefficient and tangent-stiffness proportional damping models. To this end,
Fig.1 shows response of a simple SDOF oscillator with initial period of 0.5 sec and a
thin modified Takeda hysteresis rule subjected to 10 seconds of a 1.0 Hz forcing
function. The steady-state response of the pier corresponded to a displacement ductility
of about 7.7 at the upper limit of reasonable ductile response.
Results for the stabilised loops, ignoring the transitory first three seconds of response
are plotted in Fig.1(a) (initial-stiffness proportional damping) and Fig.1(b) (tangentstiffness proportional damping). In each case the hysteretic response associated with
nonlinear structural response is plotted on the left, and the elastic damping forcedisplacement response is plotted to the right. The areas inside the loops indicate the
relative energy absorption. For the case with initial-stiffness proportional damping, the
energy absorbed by elastic damping is approximately 83% of the structural hysteretic
energy dissipation, despite the high ductility level. This might be surprising when it is
considered that the elastic damping corresponds to 5% of critical damping, while the
2000
2000
1000
1000
-1000
-2000
-1000
-2000
-0.1
Displacement (m)
0.1
-0.1
Displacement (m)
0.1
2000
2000
1000
1000
-1000
-2000
-1000
-2000
-0.1
Displacement (m)
0.1
-0.1
Displacement (m)
0.1
be seen by comparing the upper and lower right-hand plots of Fig. 1. In the lower plot,
the reduction in damping force corresponding to the stiffness change is clearly visible. In
this case, the area of the elastic damping loop is only about 15% of the structural
hysteretic energy dissipation.
Analyses of SDOF systems subjected to real earthquake records (Priestley and Grant
(2005)) show that the significance of the elastic damping model is not just limited to
steady-state response. Figure 2 shows a typical comparison of the displacement response
for SDOF oscillators with initial stiffness and tangent stiffness elastic damping. In this
example the El Centro 1940 NS record (amplitude scaled by 1.5) has been used, the initial
period was 0.5 seconds, a Takeda hysteretic rule with second slope stiffness of 5% was
adopted, and the force-reduction factor was approximately 4. The peak displacement for
the tangent- stiffness elastic damping case is 44% larger than for the initial-stiffness
damping case, indicating a very significant influence.
0.15
Tangent Stiffness
Displacement (m)
0.1
Initial Stiffness
0.05
-0.05
-0.1
0
8
12
Time (seconds)
16
20
A selection of results from a series of analyses of oscillators with different initial periods
between 0.25 and 2.0 seconds subjected to a suite of ATC32 (1996) specrrum compatible
accelerograms is shown in Fig.3, which presents the ratio of peak inelastic displacement
response to peak elastic displacement response for three different hysteresis rules, and
three different force-reduction factors, applied to the average elastic peak response force.
2.4
2.4
2.4
R=2
1.6
1.6
TS
1.2
1.2
IS
0.8
0.4
1.6
R=4
TS
0.8
1.2
1.6
R=6
IS
1.2
IS
0.8
0.8
Period (seconds)
TS
0.4
0.8
1.2
1.6
Period (seconds)
0.4
0.8
1.2
1.6
Period (seconds)
2.4
2.4
R=2
1.6
TS
1.2
0.8
1.6
IS
1.2
IS
R=4
TS
1.6
0.4
0.8
1.2
1.6
Period (seconds)
R=6
IS
1.2
0.8
TS
0.8
0.4
0.8
1.2
1.6
Period (seconds)
0.4
0.8
1.2
1.6
Period (seconds)
2.4
R=2
1.6
2.4
R=4
1.6
TS
IS
0.4
0.8
TS
1.6
IS
IS
1.2
0.8
TS
1.2
1.6
Period (seconds)
1.2
1.2
0.8
0.8
0.4
0.8
1.2
1.6
Period (seconds)
(c)Flag Hysteresis
0.4
0.8
R=6
1.2
1.6
Period (seconds)
Deck mass
fc = 39MPa
fy = 514 MPa
Mass = 7.8 tonnes
88 cm
Column
Spiral 6/6 cm
28 cm
45 cm
200 cm
288 cm
1810
Spiral 6/3 cm
Pier cross-section
88 cm
50 cm
156 cm
Spiral 6
Footing
The first test was carried out under static cyclic displacement-controlled excitation at
levels of displacement ductility of = 1, 2, 4 and 6, with three cycles at each ductility
level, followed by an elastic cycle to enable stiffness degradation to be determined.
Force [kN]
40
20
0
-0.12
-0.1
-0.08
-0.06
-0.04
-0.02
0.02
0.04
0.06
0.08
0.1
0.12
-20
-40
-60
-80
SeismoStruct simulation
Experimental test
-100
Displacement [m]
It will also be noted that the SeismoStruct simulation slightly underestimates the flexural
strength, and results in hysteresis loops that absorb more energy than the experimental
loops. This is a consequence of the difficulty in modelling shear deformation in fibreelement based analyses. Strength degradation on cycling to a specified ductility level is
also slightly under-predicted by the SeismoStruct modelling, but overall, the simulation is
excellent.
The Takeda hysteresis modelling was, as noted above, adjusted to provide the best fit to
the experimental results. Strength degradation was not included in the modelling, but
Force [kN]
40
20
0
-0.12
-0.1
-0.08
-0.06
-0.04
-0.02
0.02
0.04
0.06
0.08
0.1
0.12
-20
-40
-60
-80
Experimental test
Ruaumoko simulation
-100
Displacement [m]
The second unit was tested on a shake table using an accelerograms based on the 1984
Morgan Hill record. This record was chosen based on preliminary analyses with a number
of accelerograms where the aim was to provide the largest difference between predicted
peak displacement using initial-stiffness and tangent-stiffness elastic damping models.
The record was scaled in the time-domain as a consequence of the model scale, and also
scaled in amplitude to produce a predicted peak displacement corresponding to
displacement ductility demand of = 6, based on the tangent-stiffness prediction. This
represented the maximum stable displacement response of the static model (see Figs 5
and 6) at larger displacements strength degradation was rapid. It should be noted that
the peak predicted displacement response using initial-stiffness damping and the Takeda
simulation was of the order of = 3. The difference was larger when damping was added
to the SeismoStruct simulation, with initial-stiffness damping predicting average peak
displacements (average of positive and negative peaks) only 39% of the tangent stiffness
predictions. It was thus felt that the dynamic test could be confidently expected to
establish which damping model was more appropriate for this test.
The recorded table excitation was very close to the modified Morgan Hill record, though
peak acceleration amplitudes were about 10% higher than expected. As a consequence,
predictions from the SeismoStruct and Takeda simulations were re-run using the
10
Displacement (m)
0.1
5% T.S. Damping
0.05
5% I.S. damping
0
-0.05
-0.1
Experiment
-0.15
0
Displacement (m)
0.15
6
8
Time (sec)
(a) SeismoStruct Simulation
10
0.1
5% T.S.damping
0.05
5% I.S. Damping
12
0
-0.05
-0.1
Experiment
-0.15
0
6
Time (sec)
(b) Takeda Simulation
10
12
Figure 7 Comparison of Experiment and Predicted Displacement Time-Histories for Dynamic Test
Considering first the experimental response, it is seen that the peak displacement rises by
T = 3.5 sec. in three cycles to about 0.1m, the value corresponding to = 6 from the
11
static test, and hence the largest stable static response level. Following this the
experimental response indicates a significant residual displacement (about 0.05m) with
another large displacement pulse between T = 10 and 10.5 seconds. The negative
displacement in this pulse was 0.15m, corresponding to a displacement ductility factor of
= 9, and was accompanied by severe strength degradation. This was eventually
accompanied by fracture of all flexural reinforcement at the column base, and
catastrophic failure of the column.
The SeismoStruct prediction of response (Fig.7a) based on a tangent-stiffness elastic
damping model provides a good representation of the experimental response, particularly
up to T = 5.5 seconds. After this the residual displacement is slightly underestimated.
Beyond T = 10 seconds, the strength degradation resulting from fracture of the flexural
reinforcement was not captured in the SeismoStruct simulation, and displacements were
underestimated. Considering peak displacements up to T = 10.2 seconds, the
SeismoStruct peaks were 76% and 93% of the experimental peak positive and negative
displacements respectively. On the other hand, displacements predicted by the initialstiffness elastic damping simulation severly underestimated the peak experimental
displacements, being 27.5% and 36% of the peak positive and negative displacements.
The Takeda line-element simulations were carried out using Ruaumoko (Carr, (2005)). It
will be seen from Fig.7b that the tangent-stiffness simulation provided extremely close
estimates of the experimental response in the initial cycles up to the peak static
displacement capacity of 0.1m, which occurred at about T = 3.5 seconds, while the initialstiffness simulation underestimated the displacements by about 30% on average, except
in the initial elastic cycles (<1). Beyond T = 3.5 seconds the residual displacement was
underestimated by both initial-stiffness and tangent-stiffness models, but the peak
displacement at T = 10.2 seconds was again captured well by the tangent-stiffness model.
Considering peak displacements up to T = 10.2 seconds, the tangent-stiffness model
peaks were 104% and 102% of the experimental positive and negative displacement
peaks, while the initial-stiffness values were 55% and 67% of experimental peaks
respectively.
3. DISCUSSION OF RESULTS
The different analytical results provide compelling support for the use of tangentstiffness, rather than initial-stiffness elastic damping. It should be noted, however, that
for the fibre-element based SeismoStruct simulation, it can be argued that no elastic
damping should be included, as the nonlinearity in the elastic range of response should
be directly modelled. A further analysis was thus carried out with SeismoStruct, with no
added elastic damping. The results, which are not included in Fig.7a), were very similar,
though slightly larger, than those obtained with tangent-stiffness damping. Peak
12
displacements were increased to 91% and 98% of the experimental peak positive and
negative displacements.
For comparison, the Takeda analysis was also rerun for zero elastic damping.
Displacement peaks increased to 118% and 106% of experimental peaks.
It will be noted that the predictions for 5% initial-stiffness damping from SeismoStruct
and Ruaumoko differ significantly (compare Figs 7a and 7b). This is because the
SeismoStruct simulation used the initial un-cracked section stiffness, whereas the
Ruaumoko simulation was based on effective stiffness to first yield. A stiffness difference
by a factor of approximately 2.5 thus resulted between the two simulations, resulting in a
much higher elastic damping force for the SeismoStruct analyses. The question could
thus be asked to advocates of initial-stiffness damping: what value of initial stiffness
should be used.
4. CONCLUSIONS
Common sense considerations of structural behaviour indicate that initial-stiffness based
elastic damping is inappropriate for dynamic analysis, resulting in elastic damping forces
that are unrealistically high. Tangent-stiffness elastic damping appears to be more
appropriate, and results in increased displacements compared with initial-stiffness
predictions, particularly at high ductilities, and when hysteretic energy is low.
Results of a shake-table test of a simple bridge pier confirmed that tangent-stiffness
elastic damping gave the best prediction of displacement response when a simple Takeda
simulation, with linear response in the elastic range was used. When a more sophisticated
fibre-element simulation was used, the best simulation was obtained when no elastic
damping was used.
5. REFERENCES
Applied Technology Council (1996) ATC32: Improved Seismic Design Criteria for California Bridges
Provisional Recommendations Redwood City, CA, 256pp
Carr, A.J. (2005) Ruaumoko User Manual Dept of Civil Engineering, University of Canterbury.
Otani, S. (1981) Hysteretic Models of Reinforced Concrete for Earthquake Response Analysis
Journal of Faculty of Engineering, University of Tokyo, Vol.36(2) pp 407-441
Priestley, M.J.N. and Grant, D.N. (2005)Viscous Damping in Seismic Design and Analysis
Journal of Earthquake Engineering Vol.9 (SP2) pp 229-255
Priestley, M.J.N., Calvi, G.M., and Kowalsky, M.J. (2007) Displacement-Based Seismic Design of
Structures IUSS Press, Pavia, 721 pp.
SeismoSoft (2007) SeismoStruct A computer program for static and dynamic nonlinear analysis
of framed structures available from URL:http://www.seismosoft.com