EARTHQUAKE ENGINEERING AND STRUCTURAL DYNAMICS, VOL.

25,435453 (1996)

EFFECT OF MODELLING ASSUMPTIONS AND INPUT MOTION

CHARACTERISTICS ON SEISMIC DESIGN PARAMETERS OF RC
BRIDGE PIERS
A. S. ELNASHAP AND D. C. McCLURE'
Department of Civil Engineering, Imperial College of Science, Technology and Medicine, London SW7 2BLI. U.K.

SUMMARY
Modelling assumptions, boundary and loading conditions have a significant effect on analytical assessment of ductility
supply and demand measures for RC bridges, a structural form which had suffered extensively in recent earthquakes. In
recognition of the important role played by analysis in advancing seismic design of bridges, this paper is concerned with
assessing the effect of model characteristics and earthquake strong-motion selection on analytical action and deforma-
tion seismic design parameters. This is of particular significance when viewed in the light of the large capital investment
and problems with the satisfaction of dynamic similitude encountered in physical testing of piers and pier-deck
assemblies. The models studied range between simple fixed-base cantilever and inclusion of both soil and deck effects,
represented by assemblies of springs in translational and rotational degrees of freedom. Moreover, two sets of
earthquake records are used in dynamic analysis, each comprising six records covering low, intermediate and high
a/u, where a and u are the peak ground acceleration and velocity, respectively. The two sets differ in the scaling
procedure employed to bring them to a common level of severity; the first set is obtained by direct acceleration
scaling whilst the second utilizes the concept of velocity spectral intensity. The results from static and dynamic
analysis, using advanced material characterization and solution procedures, are assessed and discussed. Subject to the
limitations of the study, outlined in the paper, the results indicate that the inclusion of deck stiffness and/or soil
representation is essential to avail of accurate seismic response parameters. However, the effect of variations in soil
stiffness and/or deck torsional rigidity applied in the analysis is rather small, compared to the inclusion/exclusion of the
model feature. Moreover, it is also observed that using acceleration scaling leads to much larger scatter in the results than
when velocity spectral intensity scaling is used. Finally, the results from two particular earthquakes, Friuli and El Centro,
highlight the peril of using a small number of records selected without due consideration to the relationship between their
wave form, predominant periods and spectral shapes on the one hand and the response periods of the structure on the
other.

KEY WORDS: seismic design; bridges; reinforced concrete piers; ductility; strong-motion scaling

1. INTRODUCTION
Seismic performance of reinforced concrete bridges has been attracting considerable attention, following the
collapse of several structures in the San Fernando (1969),the Loma Prieta (1989) and the Northridge (1994)
earthquakes in California. Concerns have been expressed regarding the safety of such structures in other
parts of the world, in view of the belief that seismic design of bridges in California is quite advanced. This was
further emphasized by the devastation of the transportation network due to the Hyogo-ken Nanbu (Kobe,
Japan) earthquake of 17 January 1995. Notwithstanding the direct human and economic losses from the

* Professor of Earthquake Engineering

Post-graduate student

CCC 0098-8847/96/050435-29 Received I5 August I995

0 1996 by John Wiley & Sons, Ltd. Revised 12 December 1995
436 A. S. ELNASHAI AND D. C. McCLURE

heavy damage or collapse of a major bridge structure, the loss of an important transportation artery causes
great damage to business, especially in affluent societies. This, reinforced by the field observations which
indicate that there is a lot still to be learnt in seismic design of RC bridges, lends weight to the concerted effort
dedicated to studying the inelastic seismic response of such structures, not only in California, but worldwide.
Moreover, the imminent introduction of the Eurocode 8 chapter on Bridges has stimulated interest in this
subject in Europe.
To evaluate the supply of RC bridges, required to meet the demand imposed by strong ground motion,
experimentaltesting of components is required. Since small scale testing, particularly under dynamic loading,
encounters problems of similitude, full, or near-full scale testing is preferable, rendering investigations of
seismic supply of bridge components prohibitively costly. Therefore, analytical procedures, using material
and member models which incorporate the salient behavioral representations, are of great significance.
Analytical results may be verified by experimentation in a targeted and an optimally steered fashion. Issues of
demand are not discussed herein, but it suffices to state that the case for utilization of analytical methods in
demand assessment is even stronger than in supply estimation.
Having highlighted the role played by, and the potential of, analytical investigations of seismic supply of
bridges, it is important to discuss the current capabilities for such procedures. Constitutive relationships for
the two constituent materials, steel and concrete, have advanced considerably in recent years, with cyclic
degrading models and passive instantaneously varying confinement for concrete, and with multi-surface
plasticity formulations and mixed hardening rules for steel. Moreover, local reinforcement buckling and
geometric second-order effects can be adequately accounted for. In spite of this, analytical results for similar
structures vary enormously if undertaken by different research groups using different programs. A recent
example is the study of a ten-span curved RC bridge used as a bench-mark for the ‘Second International
Workshop on Seismic Design of Bridges’ held in New Zealand in August 1994. The results of earthquake
analysis of the bridge presented by Petrangeli and Pinto,’ Monti and Nuti,’ Kodera et al.,3 Kowalsky and
Priestley: Singh and Fenvess and Dodd et aL6 vary enormously. This variation is attributed to modelling
differences rather than to variations in the constitutive relationships used or the solution procedures
adopted. This is confirmed by a recent comprehensivestudy of models used to represent the cyclic response of
reinforced concrete structures undertaken by Chan and Mander.’ Whereas differences exist, they are by no
means spectacular,and would not account for the differences observed in the above-mentioned studies of the
curved bridge.
To investigate the effect of modelling assumptions, and to provide general guidelines for the accurate
modelling of RC bridges, this study focuses attention on a single pier-foundation-deck system. Various
levels of modelling rigour are used and a selected set of seismic response parameters monitored. In addition
to structural model details, aspects of input motion variation are briefly examined, in terms of the selection
criteria of a suite of earthquake records, and the method of scaling acceleration time-histories.

2. ANALYTICAL MODELLING
To investigate the response of the bridge pier using different modelling assumptions, an accurate determina-
tion of inelastic deformations in reinforced concrete is necessary. Such analysis requires material constitutive
models and accurate member representation to predict the behaviour. In this section a brief account of the
numerical element types and constitutive material models is given.

2.1. Element formulation and material models

To determine accurately member response, the advanced analysis program ADAPTIC (developed and
verified at Imperial College)was used, employing fibre elements with transformations from elemental chord
to global axes to account for large displacements. The fibre element formulation adopts a cubic shape
function and permits the spread of inelasticity across its cross section and along its length. Inelastic
behaviour is defined in terms of the selected constitutive material laws, with a check on the stress state carried
out at a series of monitoring points. The reinforcement steel material model used was the commonly cited
 SEISMIC DESIGN PARAMETERS OF RC BRIDGE PIERS 437

bilinear kinematic hardening stress strain relationship, while a uniaxial cyclic model (Reference 8) was
applied, including the significant effects that determine concrete behaviour, such as passive confinement and
cyclic degradation.
Modelling of the soil stiffness and deck stiffness was performed using joint element springs with decoupled
shear, axial and moment behaviour.

2.2. Estimation of concrete conjinement factors

Confinement has a major influence on the response of concrete in its principal direction and thus is
a significant parameter in the concrete constitutive model. Two simplified procedures were considered for the
calculation of confinement factors: first the approach set out in Eurocode 8;' and the second another
empirical relationship derived by Mander et aZ."
According to the EC8 approach, the confinement factor K, is given by

1.0 + 5.0ac-
f; fi < 0-05
for -
fco L O

1.125 + 25acL-
f; fi 2 0.05
for -
o fco

The equation given by Mander et a1." is as follows:

J
K , = - 1.125 + 2-254 1 + 7*94ac- fi
f; - 2ac-
fco fco

where

is the compressive strength of unconfined concrete,

is the confining pressure,
is the confinement effective coefficient,
is the volumetric ratio of confining steel,
is the yield strength of confining stirrups,
is the spacing of the confining stirrups and
is the diameter of the confined concrete core.

Whereas both expressions are based on the assumption of the simultaneous attainment of yield in steel and
peak stress in concrete, the use of more rational passive confinement models was not considered necessary,
since this study is aimed at establishing comparative measures of response assessment under different
modelling assumptions. Moreover, there is no clear technical argument in favour of either of the above
expressions; hence the more conservative EC8 approach was adopted.

3. RESPONSE CRITERIA AND LIMIT STATES

In order to determine seismic response parameters in both the static and dynamic analyses, it is necessary to
define a set of limit states for yield and ultimate. The limit state definition remains controversial and results
obtained from one study are not, in general, comparable to another due to the wide variation in this
definition. These are discussed hereafter, in the light of the wider discussion presented by Elnashai and
Beith.'

3.1. Yield limit state

The criteria initially considered for yield were two established definitions: first yield of the longitudinal
reinforcement; and the secant of the load deflection curve.
438 A. S. ELNASHAI A N D D. C. McCLURE

(i) First yield. First yield is defined as the point at which the main longitudinal reinforcement reaches the
material yield strain. Problems exist with this definition in terms of numerical estimates when
compared to experimentally obtained values. Generally, values observed experimentally tend to be
greater than those obtained numerically. This arises from bond slip between the longitudinal reinforce-
ment and the concrete, combined with early stiffness degradation due to cracking of the section. It was
decided that the point of first yield would not in fact be taken when the outer-most pair of
reinforcement bars yielded, as this was seen to have minor overall effect. Instead, the point of first yield
was taken to coincide with yielding of five reinforcement bar pairs. This was verified by mapping the
load-displacement curve onto the stress-strain curve of the reinforcing bars. A departure from
linearity was observed closer to the yielding of five bars than otherwise. This is clearly a function of the
particular structure considered and may require re-assessment for a different configuration.
(ii) Secant yield. A commonly employed definition of yield is used herein whereby the yield displacement
of the system is equal to that of an equivalent linearly elastic-perfectlyplastic system. The latter has an
initial stiffness given by the point of intersection of the line at 75 per cent of ultimate load with the
load-displacement curve of the inelastic system. The plastic load of the equivalent system is taken as
the ultimate load of the inelastic system.

3.2. Ultimate limit state

Even more than the yield limit state, the definition of ultimate structural response is a particularly
subjective decision. Consequently a number of ultimate criteria were considered, as follows:
(i) Concretefailure. This definition is based on an approach used by Lappas and Tassios,” who defined
the point of ultimate limit state from the maximum available curvature. This curvature is assumed to
correspond to the post-peak value of curvature at 0.85 of the peak moment, hfp& Multiplying this
value of curvature by a fictitious, but conservative, value of plastic hinge length L/4, the corresponding
maximum available rotation is obtained. In the case of the pier under investigation the axial load
remains unaltered. This implies that the curvature at failure would be constant and thus the value of
strain at 0.85 h f p e a k would also correspond to failure. Finally, the stress in the confined concrete is
related to the moment, therefore the ultimate strain could alternatively be calculated from a 15 per
cent drop in the peak stress of the confined concrete.
(ii) Signifcant drop in resistance. A commonly cited criterion that is based on prescribing the ultimate
limit state as the deformation corresponding to a drop in the load carrying capacity of the
load-deflection curve, typically in the range of 5-10 per cent.
(iii) Fracture oftensile reinforcement. This criterion defines the ultimate limit state using a limiting value of
tensile strain in the main longitudinal reinforcement. The limiting value of strain is taken as that
permitted in Eurocode 8, which is 12 per cent.
(iv) Limit on excessive rotation. An empirical procedure essentially limiting element rotation may be
utilized. Using such a definition of ultimate limit state constrains the rotational ductility to a predeter-
mined value.

4. LIMIT STATES AND RESPONSE PARAMETERS

To assess the results generated in both the monotonic and dynamic analyses a series of response parameters
must be defined. Parameters of most interest in seismic design are deflection, rotation and curvature
ductilities as well as plastic hinge length. These parameters collectively reflect the capacity of the structure of
absorb and dissipate the input earthquake energy.

4.1. Yield criteria

As discussed in Section 3.1, two yield criteria were considered, namely, first yield and secant yield. Initial
tests showed that both criteria were in close agreement for the case of fixed base models. As greater degrees of
flexibility at the pier support were introduced through modelling assumptions, a disparity became evident.
 SEISMIC DESIGN PARAMETERS OF RC BRIDGE PIERS 439

This could be attributed to greater involvement of P-A moments resulting from rigid body rotations at the
pile cap. In such cases the load-deflection curves peak at lower values and at larger displacements. This in
turn gave significantly larger values of yield displacement based on secant yield. These yield values were not
considered representative after inspection of the level of strain imposed on the main reinforcement. To avoid
such disagreement the yield criterion became solely defined using first yield.

4.2. Ultimate criteria

With regard to ultimate limit state mentioned in Section 3.1 fracture of the main reinforcement at the
ultimate material strain of 12 per cent was never achieved in initial tests. Therefore, three criteria were utilized,
namely limiting strain level of confined concrete, excessive rotation limit and significant drop in resistance.

4.3. Response parameter calculations

The adoption of one yield and three ultimate limit state definitions furnished three values of each response
parameter, for each modelling assumption. This in effect gave each parameter a band of values with an upper
and a lower bound.
(i) Class I responseparameters. This set of parameters were evaluated using first yield associated with the
ultimate limit state defined in terms of the limit on excessive member rotation. Definition of ultimate
conditions in such a manner constrains the rotational ductility. This consequently necessitates
a predetermined knowledge of the likely response for such a structure and structural material. A value
of po = 8 was assumed from previous work carried out by Elnashai and Beith,' and was subsequently
found to give rather more conservativeestimates of ultimate values. Class 1 can be considered as the
lower bound set of parameters.
(ii) Class 2 responseparameters. This class utilizes the strain corresponding to a 15 per cent drop in peak
stress of the confined concrete stress-strain relationship as the ultimate condition.
(iii) Class 3 response parameters. The final set of parameters are calculated using the significant drop in
resistance definition. A drop of 5 per cent in the load deflection response curve was chosen. This
produced the largest estimates of ultimate limit state response parameters and thus this final class
represents the upper bound.

5. STRUCTURAL MODELLING ASSUMPTIONS

The initial study of modelling assumptions was aimed at assessing the effect of varying the degree of
structural complexity. Each modelling assumption adopted is discussed in the following sections in terms of
physical nature and the subsequent numerical format. The structural models considered represent the
progressive steps in modelling complexity that might be undertaken in a design office environment.
The fundamental pier model chosen for the study was obtained from an international collaborative study
of a 3-D RC curved bridge which was presented in a workshop held in New Zealand.I3 Whereas the
workshop was concerned with the global response of bridges as a whole, the purpose of the current study was
to investigate the modelling and response of piers. Thus, a typical pier design was selected from the brief given
in Reference 13, as shown in Figure 1.
The generic pier had the following parameters:
Cylinder compressive strength feu 35 MPa
Concrete tensile strength ft 3 MPa
Characteristic steel yield f, 430 MPa
Ultimate strength (at 12 per cent strain)f. 665 MPa
Strain at yield E, 2-15x 10-3
Strain hardening parameter 1.0 per cent
Pier diameter 1500 mm
Pier height 7500 mm
440 A. S. ELNASHAI A N D D. C. McCLURE

Main longitudinal reinforcement 48D32 in pairs over the full height

Transverse reinforcement D12 @ 70 mm centres for the bottom 20 per cent of the pier
D12 @ 140 mm centres for the remainder of the height
Confinement factor K , 1.026 for the bottom 20 per cent of the pier
1.010 for the remainder of the height

1
h

III

Figure l(a). Pier and footing details.

 SEISMIC DESIGN PARAMETERS OF RC BRIDGE PIERS 441

Section A-A

140mm
Cover :SOmm lo main steel
COLUMN REINFORCEMENT

Section 8-8
-
24 024

I 1.00m

DlOrP 65m
Cover :SOmm
FOUNDA TION CYLINDER REINFORCEMENT
Figure l(b). Pier column and pile reinforcement details

The ection sh ar capacity is given by the summation of contributions from the concrete V,, steel reinforce-
ment Vs and axial force V,.
v,,, = v, + vs+ v, (3)
where

Vs= -As!scot
x 30
2 s

In these equations, AeroMis the cross-sectional area of the concrete, k reflects the degradation of shear
strength with increasing displacement ductility, D and D’ are the diameters of the pier and the confined
concrete core, L is the height of the pier and P is the axial load acting on the pier. Subject to axial force and
moments, the following equations describe the variation of k with pier displacement ductility, pa:

Application of the relevant material properties and dimensions to equations (3)-(6) gave a shear capacity
V,,, = 4855 kN.
442 A. S. ELNASHAI AND D. C. McCLURE

Finally, due to symmetry of the section, two-dimensional analysis is deemed sufficient for the purpose of
the study. Furthermore the response of multi-span bridges in the longitudinal direction has been shown to be
less significant, compared to the response in the transverse direction, Monti and NuL2

5.1. Pier model as a simple cantilever (MSI)

A control model was chosen as the simplest form that would be selected in a design office to carry out
preliminary calculations. A fixed base pier acting as a simple cantilever was therefore considered, denoted by
MS1.
Numerical modelling of the tapered section was implemented by splitting the taper into three 0 5 m
subsections, and applying the mid-height properties of each subsection to the respective element. Due to its
high flexural stiffness the taper section was unlikely to exhibit appreciable deformations. The pier column
was discretised using five elasto-plastic elements each of 1.0 m length. Finally the base node was fixed in
translation x, y and rotation about the z axis, thus forming a perfectly idealized cantilever, as shown in Figure
2. Since the tapered section was explicitly modelled, the deck weight was added at its natural location, with
no recourse to the use of rigid links.
The majority of axial load developed in the pier evolves from the 40m length of deck it supports.
Consequently a concentrated load was imposed at the pier tip to represent the deck weight, in addition to
distributed self-weights of particular importance for the rather heavy taper section.

5.2. Control model with plastic zone mesh reJinement wS2)

The first modelling assumption investigated was aimed at assessing how the level of discretization would
affect the structural response. The cubic elasto-plastic element employed in the pier mesh displays a high
degree of versatility in its ability to permit the spread of plasticity. However, it was deemed that if a coarse
mesh was used, the location of the Gauss point closest to the base may not permit an accurate assessment of
the behaviour of the most critical section in the pier.
The pier mesh was subsequently reconstructed with a degree of refinement such that base elements were
limited to 500 mm or approximately L/10. This modelling assumption would thus provide a yard stick to the
maximum distance from the base to the first integration station needed for accurate prediction of local and
global structural response.

5.3. Pier models incorporating foundations (MS3 and MS4)

The first stage of additional model complexity that might be envisaged in analysis would be to include the
influence of foundation effects. A two-stage approach was considered. Firstly to model the pile and pile cap

RC eh.clllr
/Pk- 1.b)

>
/,//,///,///&
/) /, ADAPl'ICMub

Figure 2. Modelling assumption MS (1) (control model)

 SEISMIC DESIGN PARAMETERS OF RC BRIDGE PIERS 443

but to neglect the soil and secondly to include the soil properties. The former was included in this study as an
additional case of the latter (soil stiffness sensitivity) when the soil stiffness approaches zero.

5.3.1. Pier model with foundation structure - no soil eflects MS(3). Details of the pile group design were
adopted from previous studies,I3 as shown in Figure 1. The piles are of drilled type and extend to a depth of
5 m in cohesionless sand, at which point they are assumed to end bear on bed rock. Pile group reinforcement
is given below:

pile
Main longitudinal reinforcement 24D24 over the full length
Transverse reinforcement DlO spirls with 65 mm pitch for the full height
pile cap
Main reinforcement D24 @ 200 mm centres each way top and bottom
Transverse reinforcement D16 @ 400 mm centres each way
On the assumption that the piles are end-bearing, pinned supports were used to represent the socketing
connection made in rock. Although the provision of pinned supports would prevent vertical freedom at the
pile base, a situation which would not be achieved in reality, preliminary tests showed that no uplift
developed at the rock-pile interface. With the possibility that inelastic behaviour could ensue in the
piles, elasto-plastic elements were utilized. Due to the size of the pile cap and level of reinforcement
specified, it was deemed that only elastic deformations would prevail in this section. Thus to reduce any
unnecessary computational effort, the pile cap was idealized using a solid steel cross section with a structural
stiffness identical to that of the reinforced concrete pile cap. A diagrammatic representation is shown in
Figure 3.

5.3.2. Pier model with foundation structure - including soil (MS4). Having considered the extreme case of
including the influence of foundation structure but neglecting the soil, a further series of models were set up to
assess the sensitivity of parameter response to soil stiffness. The value of lateral subgrade coefficient quoted in

Figure 3. Modelling assumption MS (3) (pier incorporating foundation structures, nil soil stiffness)
444 A. S. ELNASHAI AND D. C. McCLURE

Reference 13 was selected as a benchmark for typical soil stiffness of cohesionless sand. By multiplying and
dividing this value using a factor of two, a range of values were generated to represent the limits of possible
soil stiffness.
Whereas modelling the soil may be performed using continuum finite and boundary element procedures,
within the context of structural analysis, spring representation is sufficient. This is the approach adopted
here, after considering more complex mechanical analogue models available in the literature. For lateral
loading the spring compliances are usually referred to as p-y curves. As a first approximation, the soil
response was considered linear. To arrive at the value of soil spring stiffness the value of lateral subgrade
coefficient given in Reference 13, which is a function of depth, was calculated as a piece-wise constant
function of depth, with a step equal to the pile element mesh spacing. The total force resulting from the
continuous soil reaction was modelled as a discrete spring force. The springs have equal tension and
compression characteristics. Since the piles are end-bearing, no vertical springs were used for skin friction
representation. Modelling of vertical stiffness of the soil located underneath the pile cap was neglected, on the
premise that this material would be back fill of negligible strength. The model used in shown in Figure 4. In
this analysis, the mass of the soil as well as its radiation damping characteristics were not included. This
decision was taken since very few practical analyses of bridges go that far in complexity and the ensuing
number of parameters (accounting for soil properties other than stiffness and strength) would have been
prohibitively large. Finally, no group effects were taken into consideration, for the latter reasons.
The models set up to represent the range of soil stiffnesses were denoted as follows: MS4a low soil stiffness,
MS4b medium soil stiffness and MS4c high soil stiffness.

5.4. Pier incorporating foundation and deck injluence (MS6)

As a further level of model elaboration that might be contemplated in analysis, a provision was made for
possible deck stiffness that could be mobilized due to the type of pier-deck support condition. Under pier
loading transverse to the plane of the deck, support bearings could generate a certain degree of torsional
moment. An upper bound value of available moment would exist when the entire deck load would be
transferred to one bearing. Assuming a spacing of bearings from the pier centre line of 3 3 m, this would
correspond to a moment of 19.4 M Nm, based on the dead and live loads on the deck acting equally at the

I 1
Jhthquake lnput applied
at dl support poinb
afmfirp*

Soil springs modelbg

in a pieambe manner
the varltlon of lateral
subgrade stIfKness wlth depth

Figure 4. Modelling assumption MS (4) (pier incorporating foundation structure and soil stiffness)
 SEISMIC DESIGN PARAMETERS OF RC BRIDGE PIERS 445

Figure 5. Modelling assumption MS (6) (pier incorporating foundation structure, soil stiffness and deck restraint)

two bearing support points. In the absence of any data with regard to the performance of bearings, it was
assumed that they would possess adequate rigidity such that the full deck torsional stiffness would be
developed. A joint element, similar to that used for soil property modelling, was employed with zero axial and
shear component but with a rotational stiffness corresponding to that of the deck. Moment-
rotation characteristics for the element were implemented using a bilinear curve with initial stiffness
equal to the torsional stiffness of the deck up to the point of maximum available moment given above.
Beyond this, zero stiffness gradient was defined to limit the maximum developed moment. Consider-
ing a typical pier from the literature (Reference 13), the torsional constant J was calculated as 11-7m4, whilst
the torsional stiffness ke was 22.44 x 10” Nm/rad. A graphical representation of the model is shown in
Figure 5.
In a manner similar to soil stiffness, three models of deck stiffness were set up to assess the sensitivity of the
response parameters to the deck torsional restraint. However, due to the nature of hollow sections where
stiffness vary dramatically with minor changes in cross section dimensions, a factor of five was chosen for
scaling the stiffness to both magnify and reduce the above value.
The mid-range value of lateral subgrade coefficient was used for the case of models incorporating deck
stiffnessesand foundations. The models thus generated were denoted MS6a in the case of low deck stiffness,
MS6b in the case of medium deck stiffness and MS6c in the case of high deck stiffness. Model MS5 includes
deck restraint but no soil-foundation system (justified in Section 6.2).

6. RESULTS FOR MONOTONIC LOADING

Results discussed hereafter are divided into two categories,namely, models including deck torsional stiffness
and free cantilevers, since the mode of response of the two systems is fundamentally different. Whereas
models including deck stiffness behave as shear frames, those possessing no deck stiffness respond as simple
cantilevers.
446 A. S. ELNASHAI AND D. C. McCLURE

6.1. Models possessing no deck stiflness

6.1.1. Control model -simple cantileverpier (MSI). Results of the pier acting as a pure cantilever form the
basis for comparison with the other model assumptions in this category. Moreover, with reference to Section
4.1.3, Class 1 response parameters represent the lower bound, Class 2 the mid-range and Class 3 the upper
bound values. The results from the static analysis are summarized in Tables I-IV and presented in load
deflection curves as shown in Figures 6-9.
Displacement ductility results display a range of values in which 4.2 Q pa < 6-3. Taking the lower bound
value, this compares favourably with the conservative values quoted in codes of practice for reinforced
concrete structures. Values calculated for rotation and curvature ductility also echo values obtained by
Elnashai and Beith" in a similar investigation on ductility capacity of soild circular RC piers. Little variation
occurs for values of plastic hinge length calculated using either class of ultimate criteria, and these too agree
with values quoted by Elnashai and Beith."

6.1.2. Control model with plastic zone mesh rejinement (MS2). Initial observations of the refined mesh
model result suggested little change in response with the load deflection curve of Figure 6 matching nearly

Table I. Displacements at yield and ultimate limit state. Structural state at yield and ultimate conditions
~

Yield Ultimate, Class 1 Ultimate, Class 2 Ultimate, Class 3

8 8 8 8
A (radx d, A (radx 4 A radx 4 A (radx 4
Model (mm) 1O-j) (xlO-j) (mm) ( x ~ O - ~(mm)
) ( x ~ O - ~ )(mm) lo-') ( x ~ O - ~ )

MS1 49 2.760 2.884 205 2.208 2350 220 2440 2.590 310 3.687 3.990
MS2 47 1.396 2.852 150 1-112 2360 160 1.266 2.557 370 3.840 8.860
MS3 102 2.754 2.875 310 2.200 2.400 318 2.430 2601 400 3.600 3.880
MS4a 86 2.764 2.889 261 2.211 2365 276 2.425 2-590 360 3580 3.860
MS4b 77 2.742 2.871 247 2.194 2356 263 2.425 2.594 340 3.650 3-900
MS4c 70 2-770 2.890 240 2.216 2330 260 2.460 2.630 330 3.700 3-950
MS5 15 2.560 2.921 90 2.050 2550 81 1.805 2.258 260 5.760 8.645
MS6a 146 2.572 2.927 365 2-050 2540 358 1.836 2.285 580 6.918 11.096
MS6b 110 2.550 2.921 345 2.050 2560 330 1.810 2.278 580 7.463 11.950
MS6c 100 2.552 2.890 340 2.040 2500 325 1.845 2.237 570 7.220 11.690

Table 11. Ductility parameters for classes 1-3. Response parameters

PA Pe P+

Model Class 1 Class 2 Class 3 Class 1 Class 2 Class 3 Class 1 Class 2 Class 3

MS1 4.2 4.5 6-3 8-0 8.8 13.4 8.1 9.0 13.8
MS2 3.2 3.4 7.9 8.0 9-0 27.5 8.3 9.0 31.1
MS3 3.0 3.1 3.9 8.0 8-8 13.1 8-2 9.0 13.5
MS4a 3.0 3.2 4.2 8.0 8-8 13.0 8.2 9.0 13.4
MS4b 3.2 3.4 4-4 8.0 8.8 13-3 8-2 9.0 13-6
MS4c 3.4 3.7 4.7 8-0 8.9 13.4 8.1 9-1 13.7
MS5 6.0 5-4 17.3 8-0 7.1 22.5 8.8 7.7 29.6
MS6a 2.5 2-5 4.0 8.0 7.1 26.9 8.7 7.8 37.9
MS6b 3-1 3-0 5.3 8.0 7.1 28-4 8.8 7.8 40-6
MS6c 3.4 3.3 57 8.0 7.2 28-3 8.8 7.7 41.0
 SEISMIC DESIGN PARAMETERS OF RC BRIDGE PIERS 447
Table 111. Yield and ultimate loads. Yield and collapse loading
~ ~~

Collapse force Collapse moment

Yield or” Yield (kN m)
force moment
Model (kN) Class 1 Class 2 Class 3 (kNm) Class 1 Class 2 Class 3
~

MS1 1483 1862 1853 1792 11460 15400 15420 15 500

MS2 1450 1872 1870 1781 11500 14900 14970 15 600
MS3 1468 1834 1830 1790 11440 15400 15400 15600
MS4a 1474 1845 1840 1780 11400 15400 15400 15600
MS4b 1470 1839 1850 1787 11 500 15400 15400 15600
MS4c 1475 1849 1850 1782 11500 15400 15400 15600
MS5 4740 6850 6840 6510 11 500 17500 17400 17290
MS6a 4650 6727 6713 6411 12400 17400 17400 17230
MS6b 4220 6691 6650 6384 12300 17400 17300 17200
MS6c 4068 6698 6647 6381 12130 17400 17350 17200

Table IV. Plastic hinge lengths and energy absorbed. Plastic hinge
lengths

Normalized hinge length

Absorbed hinge
Model Class 1 Class 2 Class 3 energy (kNm)

MS1 0-257 0.256 0.261 317 103

MS2 0232 0228 0263 165881
MS3 0.257 0.257 0.262 316 562
MS4a 0255 0255 0265 315 252
MS4b 0.260 0.260 0.269 315 812
MS4c 0253 0253 0263 315920
MS5 0-282 0.286 0-289 227 203
MS6a 0-287 0.287 0-291 227 193
MS6b 0-289 0293 0.295 227 753
MS6c 028 1 0.285 0.288 228 644

2ooo
1800
1600
1400

600
400

0 ,......... . . . . . I .........,......... I
0 100 m 300 400
Displacement (mm)
Figure 6. Load deflection curve for control and mesh refined models
448 A. S. ELNASHAI AND D. C. McCLURE

I I I I
I- ContmlModcl(MS1)

I ----
-.-. No Soii S W a t r s (Ma)

0 100 200 m 400

Displacement (mm)
Figure 7. Load deflection curve for control model and models possessing foundation structure

I ---. CoatmlModcl(MS1) I

0
Displacement (mm)
Figure 8. Load deflection curve for control model with and without deck stiffness

exactly the curve obtained for the control model. However, on calculation of the response parameters some
variations did emerge. Class 1 and 2 displayed values of rotational and curvature ductility consistent with
those of the control. This should be expected as the failure criteria for both these classes are based on rotation
and strain. In the absence of any change in axial load these parameters should remain the same. Displace-
ment ductility values, on the other hand, suffered a 25 per cent drop. This reduction obtained for pA can be
seen to result from more accurate modelling of inelasticity at the critical section, in turn causing earlier
achievement of failure criteria. Referring to Table I values of displacement at yield are in close agreement,
taking the mid-range values of 49 mm for the control model and 47 mm for the refined mesh model. In
contrast, the levels of displacement at ultimate display large variation being 220 mm for the control model as
compared to 160 mm for the refined mesh model. Furthermore, as the plastic hinge length is a function of
both pA and pd, a commensurate reduction is observed in the plastic hinge length.
 SEISMIC DESIGN PARAMETERS OF RC BRIDGE PIERS 449

Displacement (mm)
Figure 9. Load deflection curve for control model with deck stiffness and pier incorporating foundation and deck influence

With regard to the Class 3 response parameters, a significant increase of all three ductilities occurred. A 25
per cent increase was recorded for pa while values of and p4doubled. Such levels of increase can take place
for sectional parameters and yet only cause minor variations in the overall p A because of the short length of
element over which they take place. All of these increases nonetheless can be attributed to the milder dip
which occurs beyond the peak lateral force in the load deflection curve. Since Class 3 is based on a significant
drop in resistance, the displacement values at failure are larger, with reference to Table I.
The dramatic change in results which arise upon mesh refinement in the plastic hinge zone indicate the
sensitivity of response to the location of the Gauss point in the critical section.

6.1.3. Pier model with foundation structure - no soil eflects (MS3). Values of sectional and member
response parameters po and p4remain constant with their counterparts calculated in the control model. This,
for reasons discussed in the previous section, should be expected. Class 3 response parameters, based on
global response, once again showed a degree of variability with marginally lower values of pe and p4,as given
in Table 11. A reduced value of peak lateral force Ppeak was recorded. This does not represent a drop in the
capacity of the pier, but in fact indicates the greater involvement of P-A moments. This can be verified by
checking the moment versus displacement curves for the control and model assumption. Both reach the same
peak moment corresponding to the onset of strain hardening in the steel, but at different levels of
displacements, as shown in Figure 10.
Displacement ductility suffers a 30 per cent reduction. This variation arises not from a change in
performance of the actual pier but from the definition of pA. Due to a lower lateral stiffness of the overall
structural system, considerable displacement and rotation of the piles and pile cap takes place before the
lateral force required to cause yield is achieved. More than one-third of the displacement up to the point of
yield occurs as elastic translation of the pile cap, with a further 15 per cent of tip displacement originating
from rotation of the pile cap. Upon yield the stiffness of the pier progressively softens and its plastic
deformations begin to form the significant component of the overall displacement.As a consequence of yield
displacement being dominated by translation of the piles and rotation of the pile cap, reduced values of p A
occur in all classes. Furthermore a comparison of energy absorbed in the plastic hinge shows consistency
with the control model, as shown in Table IV.

6.1.4. Pier model withfoundation structure and soil (MS4). For the same reason of reduced lateral stiffness
each of the models with varying soil stiffness exhibits values of pA lower than the control model. The inclusion
450 A. S. ELNASHAI AND D. C. McCLURE

P k witb
FOWM~~~OII -NO
soil SIimDcrr

0 100mulo400Mo600
Displacement (mm)
Figure 10. Moment deflection curve for control model and pier incorporating foundation structure without soil stiffness

of soil, on the other hand, increaseslateral stiffness when compared to the previous system. This has the effect
of reducing elastic displacements at foundation level which occur up to yield, thereby marginally increasing
the values of pa with increasing stiffness. No appreciable change occurs in the values obtained for plastic
hinge length.
Comparison of the family of load-deflection of curves for varying soil stiffness shows little variation of
response in real terms. Over the full range of what may be considered possible sub-soil conditions only a 15
per cent variation of pa is observed. This value represents a minor degree of change when compared to the
range of difference in soil stiffness used.

6.2. Deck stiflness

Response of systems which include deck stiffness changes from that of a cantilever to a more rigid shear
frame behaviour, thus significantly altering the overall structural response. This is due to the pier deforming
in double curvature, which allows for two possible plastic hinges to develop, changing the energy absorption
characteristics. It was therefore deemed that an additional control model (MS5), a cantilever pier incorporat-
ing deck stiffness, would be necessary for purposes of comparison.

6.2.1. Cantilever model incorporating deck stifness (MS5). The first interesting feature of the results
obtained is the reversal of Classes 1 and 2 in terms of bounding values. In the previous analyses failure for
Class 1was set to a predetermined value of rotational ductility = 8.0.This value was chosen on the basis of
previous work carried out by Elnashai and Beith" on pier ductilites and formed the lower bound values of
response parameters. With the change in the pattern of structural response, this value of capacity
represents a value higher than the demand imposed by defining failure in terms of an ultimate strain.
In general, curvature ductility calculated for this control model, and subsequently for the other models
incorporating deck stiffness, display a 15 per cent reduction in Class 1 when compared to their counterparts
behaving as cantilevers. The explanation for this lies with the critical plastic hinge. While responding in
double curvature the moments generated tend to develop two plastic hinges, one at the base and the other
just below the taper section. Two levels of confinement exist within the pier column: D12 @ 70 mm centres
for the bottom 20 per cent of the column; and D12 @ 140mm centres for the remainder of the height; hence
the top hinge has lower confinement. As the foundations possess a degree of flexibility, more severe
curvatures are imposed at the constrained upper hinge location. Lower confinement gives rise to less stress
for a given strain and consequently a greater proportion of the cross section in compression. Yield is defined
 SEISMIC DESIGN PARAMETERS OF RC BRIDGE PIERS 45 1

in terms of steel strain, thus larger curvatures ensue at yield, while the ultimate limit state defined in terms of
concrete strain will produce lower ultimate curvatures, refer to Table I. The outcome reduces the curvature
ductility from p+ = 9.9 to p+ = 7.8 (Table 11).
Finally the level of displacement ductility achieved increased by 20 per cent. Inspection of the normalized
plastic hinge length however indicates an increase from z = 2-6-2-9 L. Rotation of the plastic hinge is the
integration of the curvatures over its length. Smaller curvatures result at failure, but this is offset by increased
length of the plastic hinge leading to greater displacement ductility.

6.2.2. Pier incorporating foundation and deck infuence (MS6). The values calculated for and p+
remain consistent for each of the models possessing deck stiffness. However, a significantly larger drop
occurs in displacement ductility than witnessed for the case of the pure cantilever when modified to include
foundation effects. Again this is attributed to the definition of displacement ductility. Large values of
displacement occur in order to develop the high degree of lateral load needed to produce yield. In this
case the stiffness of the systems will approach that of a shear frame, whereas the cantilever has only
25 per cent of the shear frame stiffness.Thus even larger displacements develop in the soil before yield is
achieved. Beyond yield, the pier stiffness drops and displacements become dominanted by the plastic
response of the pier. Although the control model displayed a relatively high level of displacement ductility
due to increased plastic hinge length, this is offset by response up to yield in the models including
foundations.
Several interesting observations emanate from the deck stiffness variation models. The family of
load-deflection curves obtained for variation in deck stiffness, as shown in Figure 9 display only
minor changes for such a large difference in torsional stiffness. Also, the change of response pattern, from
cantilever to shear frame, has important implications on location of critical sections where careful detailing is
called for.

6.3. Summary and results

Two distinct categories exist in the results obtained. Within each category of model, with and without deck
stiffness, three values are quoted for each response parameter, based on different criteria for the ultimate limit
state. For those models displaying cantilever response Class 1 represents the lower bound, Class 2 the
intermediate and Class 3 the upper bound. Models behaving as shear frames, however, have Class 2 as their
lower bound, while Class 1 represents intermediate values.
Mesh refinement gave a structural response in nature very similar to the control case, but inelasticity was
predicted at an earlier stage of loading, displacement ductility suffered a 25 per cent reduction in conse-
quence. The stiffer fixed base control models in both categories tend to display levels of pA higher than their
counterparts with soil-structure interaction. This results from the proportionally larger yield displacements
due to flexibility of the ground conditions. Values obtained for and p+ represent member and sec-
tional parameters and in the absence of change in axial load they remain constant independent of
modelling assumptions within each category. However, variation of these values did occur between catego-
ries and is attributed to the relocation of critical zones of inelasticity. Models possessing deck stiffness deform
in double curvature and consequently develop two plastic hinges. The hinge which forms beneath the
taper section has less confinement steel and consequently dictates failure, with lower values of po and p+
ensuing.

7. EFFECT O F GROUND MOTION CHARACTERISTICS

Seismic analysis results are affected by both the characteristics of the structural model and the earthquake
acceleration time histories used. This section concentrates on the evaluation of the effect of the characteristics
of natural earthquake records on seismic demand assessment. Furthermore, the effect of scaling of earth-
quake records, required for seismic demand assessment and the evaluation of the yield and collapse
earthquakes, is investigated.
452 A. S. ELNASHAI AND D. C. McCLURE

7.1. Structural model

For the dynamic analyses performed in the current section the most representative structural model was
selected. To this end the model incorporating soil interaction but neglecting deck stiffness was chosen. Deck
stiffness was omitted on the grounds that the pier design was not intended to mobilize any significant level of
torsional moment. Evidence for this was discussed in Section 6.2.2 and confirmed by the analysis conducted
by Monti and Nuti,' which showed that uplift at the deck-pier interface causes the pier to behave as
a cantilever.
Linearly distributed mass was employed for all structural elements; piles, pile cap and the pier including
a concentrated mass applied at the cantilever tip to account for the significant deck inertia. The basic
Newmark two-parameter algorithm was utilized for the integration of the equations of motion. This was
considered preferable to the ADAPTIC facility of algorithmicly-dampedprocedure which tends to reduce the
force amplitude. For the model incorporating the pier and foundations a time step of tstcp x 0.125 s would
have been sufficient, since the fundamental period of the structure is 1.25 s. However, the earthquake records
used in the subsequent analyses were digitized at time intervals of 0-01s. An integration time step of 0.02 s
was chosen to preserve the earthquake signal, and to obtain an accurate solution. [It is noteworthy that the
program used, ADAPTIC, applied automatic re-adjustment of time-step according to the convergence
characteristics of the analysis].

7.2. Selection of earthquake ground motion input

Where the response of a structural form to general earthquake loading is being investigated it is vital to
consider a wide range of earthquake records of varying characteristics, such as predominant period,
frequency content and duration.
Zhu et a l l 4 define three categories of earthquake ground motion: (a) normal ground motions exhibiting
significant energy over a broad range of frequencies , (b) ground motion accelerograms possessing many
large-amplitude,high-frequency oscillations, and finally (c) those records in which the significant energy is
contained in a few long-duration acceleration pulses. These characteristics can be ascribed to variations in
local soil conditions, epicentral distance and the magnitude and duration of event.

7.2.1. Concept of an a/v ratio. In the above study, it was proposed to use a 'peak ground acceleration to
peak ground velocity' ratio as a simple but representative means of identifying characteristics of earthquake
accelerograms. It is known that peak accelerations are associated with high frequencies while peak ground
velocities are linked to moderate to low frequency oscillations. Thus, ground motions of the above type (b)
will tend to display high a/v ratios, whilst for long-duration acceleration pulses of type (c) a low a/v ratio will
prevail. Accelerograms recorded near the source of an earthquake will possess high a/u ratios, while with
increasing distance the ratio will reduce. Furthermore, structures founded on rock and firm soil conditions
will experience shorter-duration high-frequency excitation than those erected on softer material.
In essence a suite of accelerogramspossessing a range of u/o ratios will not only reflect the range of possible
soil types, with the implications on response amplification, but will also include significant seismological
features likely to influence structural response.

7.2.2. Accelerograms chosen for dynamic analyses. Based on the discussion presented in the previous
section earthquake records were chosen on the principle of accelerogram a/v ratios to achieve an ensemble
which would envelope typical elastic response spectra. Three categories of a/u ratio were considered with the
ranges as specified in the National Building Code of Canada:15

low a/v < 0.8 g/ms - ' (84

normal 0.8 g/ms-' < a/v < 1-2g/ms-' (8b)

 SEISMIC DESIGN PARAMETERS OF RC BRIDGE PIERS 453

Table V. Ground motion records employed in dynamic analysis. Ground motion records
Component
Record label Earthquake event Recording station direction
Friuli Friuli Tolmezzo, EW
(6 May 1976) Italy
Gazli Gazli Karakyr Point EW
(17 May 1976) Uzbekistan
Loma Prieta EW Loma Prieta Emeryville S80W
(17 October 1979) California USA
El Centro Imperial Valley El Centro SOOE
(18 May 1940) California USA
Spitak Spitak Gukasyan Translational
(17 December 1988) Armenia
Loma Prieta NS Loma Prieta Emeryville NlOW
(17 October 1979) California USA

A suite consisting of six earthquake records was used, with two records selected from each category of a/v
ratio. The location and dates of the events are given in Table V along with the location and direction of the
recording. This set of records was used successfully by BroderickI6 for the evaluation of behaviour factors for
composite frames.
The selection consists of three Californian records, including the commonly used North-South component
of the Imperial Valley earthquake recorded at El Centro in 1940, and three Eurasian records. Both the
records from Emeryville during the Loma Prieta earthquake of 1989 are significantly different from the
majority of records from this event as they were recorded on a site of very soft soil. Of the Eurasian events, the
Gazli earthquake possesses the largest local magnitude and peak ground accelerations of the set. The Spitak
accelerogram was recorded during the destructive Armenian earthquake of 1988, and finally the Friuli record
is included on account of its very high a/o ratio and unusual acceleration time-history, exhibiting a single
high amplitude pulse.
Further details of the properties of the recording site and peak ground motions for all six records are given
in Tables VI and VII.

7.3. Variability of response due to earthquake characteristics

7.3.1. Structural behaviourfactors. For a particular response period the behaviour factor q for a structure,
as defined in Eurocode 8,9 is the ratio of the ordinates of the elastic acceleration spectrum used to define the
seismic hazard at a site to those of the inelastic spectrum employed in the derivation of the seismic design
forces:

Table VI. Site properties of recording station and local magnitude. Recording
station and magnitude of event
Epicentral
distance Magnitude
Record label (km) Soil type M L

Friuli 52 Rock 6-4

Gazli 14 Intermediate stiffness 7.3
Loma Prieta EW 97 Soft 7.1
El Centro 8 Stiff 66
Spitak 27 Intermediate stiffness 6.8
Loma Prieta NS 97 Soft 7.1
454 A. S. ELNASHAI AND D. C. McCLURE

Table VII. Ground motion characteristicsof accelerograms.Ground motion record properties

Peak ground Peak ground ratio
a/v Period of max
Record label acceleration (g) velocity (m/s) wms- l) amplificaion (s)
Friuli 0159 0080 1-99 095
Gazli 0.724 0606 1.20 013
L. Prieta EW 0.213 0216 0-99 065
El Centro 0.344 0365 094 026
Spitak 0182 0237 077 0.36
L. Prieta NS 0.250 0438 0.58 1.20

This assumes constant amplification throughout the period range. In the above, (Sa)ddenotes the design
spectral acceleration and the superscripts el and in refer to the elastic and inelastic values, respectively. In
Eurocode 8,9 maximum allowable q-factors are specified for a range of structural forms and construction
materials, whilst the same role is played by the parameter R (or R , for working stress codes) in U.S.practice.
These values are intended to reflect the ability of a structural type to undergo stable oscillations in the
inelastic range. Furthermore, they are also intended to represent the lower bounds on actual ductility supply
of individual structures. Thus it should be expected that qcodc < q', where q' represents the ultimate value not
from design spectra but actual response spectra obtained from ground motion intensities which produce
yield and collapse, i.e.

where subscripts 'c' and 'y' refer to collapse and yield, respectively.
Using the rationale that a perfectly designed structure will yield under the design forces corresponding to
equation (10) may be re-written as

a definition utilized by Kappos." While accounting for variability in the collapse earthquake such a 'defini-
tion does not account for variability in ground acceleration at yield, instead adopting the design yield value.
Referring back to equation (lo), actual structural behaviour factor is defined as

q' = ag (collapse)

ag (yield)

assuming a constant dynamic magnification factor between collapse and yield spectra. This quantity
accounts for variability of both the yield and collapse values. Subsequent behaviour factor values in this
study are based on the definition of equation (12). It is noteworthy that alternative definitions exist, the
discussion of which is beyond the scope of this work.
The approach used to calculate q factors was outlined in equation (12) and involves determining the
respective values of ground motion which cause yield and collapse. Due to prohibitive times involved for
computational analysis, only one class of ultimate definition was employed. The failure strain of concrete was
selected for this purpose, since it represents a reasonable criterion for bridge piers, as opposed to interstorey
drift often used for building frames.
Determining the values of yield and collapse earthquake can only be approached in an iterative manner, by
successively scaling the earthquake record. Here, the direction acceleration scaling procedure was utilized,
a decision further discussed in Section 7.3.2.1.
The values of behaviour factor evaluated for the six earthquakes, given in Table VIII show the extent to
which ground motion input may cause variation due to predominant frequency, frequency content, duration
and number of cycles corresponding to the particular frequencies of structural significance. The results given
 SEISMIC DESIGN PARAMETERS OF RC BRIDGE PIERS 455

Table VIII. Individual behaviour factors for each earthquake record. Behaviour factors
Behaviour factor Ratio to mean
ap fcoll.pss) 41
Yield PGA Ultimate PGA -
Record label (yield) (collapse) a, (yield) 4mean

Friuli 2.500 10.00 4.0 0.909

Gazli 0.305 1.010 3.3 0.75
Lorna Prieta EW 0.160 0-570 3.6 0.819
El Centro 0.130 0.930 7.2 1.636
Spitak 0.130 0665 5.1 1.159
Lorna Prieta NS 0.100 0.320 3.2 0.727
Mean - - 4.4 -

in Table VIII, highlight the danger of using natural earthquakes in seismic assessment studies when the suite
of records is not carefully selected. For instance, if Friuli., El Centro and the Spitak records were chosen,
a mean q factor would have been 5.4. On the other hand, the three remaining records, if used on their own,
would have indicated a mean behaviour factor of 3.3, more than 35 per cent lower than the former value.
Moreover, the results show that El Centro, which is by far the most widely used earthquake record, is by no
means of conservative loading scenario. Furthermore, the results for Friuli indicate the danger of applying
equation (11)for behaviour factor calculation, where the yield acceleration is taken as the design acceleration.
This particular record has a very narrow band of frequencies associated with high amplifications,hence all
analyses using this record give rather spurious results. This was also reported by Broderick.16

7.3.2. Variabilitydue to earthquake record scalingprocedure. To investigate the effect of ground motion as
an input parameter on demand imposed it is imperative that comparison is made with records possessing the
same level of intensity. This necessity stems from the fact that all important features of seismic response, such
as strength and ductility are highly dependent on ground motion intensity. Having ensured this, other
features such as frequency content and duration of loading can then be examined.
Earthquake records display wide variations in intensity, thus a procedure must be employed to scale
records to a common level. The effects on structural response of various scaling techniques are discussed in
the following section.

7.3.2.1. Effect of ground motion scaling on seismic response

For the purpose of calculating behaviour factors, direct acceleration scaling has the advantage of
simplicity in application and is consistent with methods adopted in design codes to define seismic loads.
However, it has long been recognized (for example by Blume et a1.18)that depending on structural period, the
response may be sensitive to ground acceleration, velocity or displacement.Three ranges of structural period
have been defined, corresponding to sensitivity to ground acceleration, velocity and displacement; T < 0.5 s,
0.5 Q T Q 3.0 s and T > 3.0 s, respectively. Short period structures will consequently be sensitive to peak
ground acceleration, while structures of moderate to long period will be sensitive to peak ground velocity.
Spectral dispersion can thus be reduced for short period structures by scaling to acceleration, with similar
reductions for longer periods by scaling to velocity. Such concepts have been recently re-confirmed by both
Tso et ~ 1 . and
' ~ Chandler.20
It would seem appropriate therefore to scale earthquake records in a manner which reflects the significant
response periods of the structure under consideration. In the case of the pier in this study, scaling to peak
ground velocity would thus be implied, in recognition of the moderately long natural period. Such a
scaling procedure, however, lacks consistency with most current design code methodology and would
not satisfy the equivalence between records and the design spectra. To overcome this difficulty it is
possible to use acceleration scaling alongside the concept of spectral intensity proposed by Housner.21The
latter is defined as the area under the pseudo-velocity spectrum curve between the periods of 0.1 and 2-5 s,
456 A. S. ELNASHAI AND D. C. McCLURE

expressed as

where represents the fraction of critical damping, S, the pseudo-spectral velocity and T the response
period. The pseudo-spectral velocity is obtained from the integration with respect to period of the acceler-
ation response spectrum through the following equation:

where S, is the spectral acceleration.

The effect of scaling to equal spectrum intensities ensures that the earthquake records possess equal energy
contents between the periods 0.1 and 2.5 s. Nau and HallZZdemonstrated that such a procedure significantly
reduces spectral dispersion for the period range under consideration. It is, however, important to state that
other intensity definitions, such as the Arias intensity, may be used for scaling purposes.

7.3.2.2. Scaling of selected ground motions

In order to compare seismic demand imposed by each ground motion record, two scaling procedures were
employed. The first was the simple approach adopted in normal design code practice, where peak acceler-
ation is linearly scaled to the target value, 0.25 g for this study. The second procedure applied was scaling of
each record to contain the same level of spectral intensity as the design spectrum with peak ground
accelerati.on0.25 g. This was achieved by firstly assessing the spectral intensity contained in each record, after
linear scaling of peak ground acceleration to the design peak value of 0.25 g . Each accelerogram was then
re-scaled, by its ratio of spectral intensity to the design code intensity, as indicated in Table IX.

7.3.2.3. Structural response

The results of the analysis under the two earthquake scaling procedures are presented in Tables X and XI.
The displacement response time histories are given in Figures 11-16, where differences in peaks, their time of
occurrence and response periods are clearly indicated; the latter is a consequence of differences in instan-
taneous inelastic stiffness. Use was made of the static results to construct relationships between concrete
compressive strain and the ductility demand, hence the results of the analysis were used directly to assess the
latter quantity.
The results further demonstrate the variability of structural response, both in terms of physical response
and ductility demand. It would appear that due to its unusual single pulse nature, results obtained for the
Friuli record exist rather as an anomaly, and were subsequently discounted from the results set for further

Table IX. Scaling of accelerograms to design peak ground acceleration and EC8 design
spectrum spectral intensity. Scaling of ground motion accelerations

Friuli 0.159 1.56 0.250 143.7 1.20 0301

Gazli 0.724 0-35 0.250 71.3 2-42 0.608
L. Prieta EW 0.213 1.17 0250 1163 1.48 0.371
El Centro 0.344 0.73 0.250 100.5 1.72 0430
Spitak 0.182 1.30 0.250 95.6 1430 0.500
L. Prieta NS 0.250 1-00 0.250 192.3 0.90 0.225
 SEISMIC DESIGN PARAMETERS OF RC BRIDGE PIERS 457

Table X. Structural response for PGA and SI scaling procedures.

Structural response

Friuli 15 0.650 18 0-750

Gazli 50 1.200 115 2.750
L. Prieta EW 180 5.000 250 9.250
El Centro 130 3.000 160 4.750
Spitak 150 3.100 210 6.750
L. Prieta NS 210 7.500 210 7.000
Mean, p 144 3.96 189 6.10
Std. dev., a 52.3 2-14 46.7 2.20
cov = u/p 036 0.54 025 0.36

Table XI. Response parameters for PGA and SI scaling procedures. Response para-
meters

Record label PA (demand pga) & (demand pga) PA (demand SI) pq5 (demand SI)

Friuli 0.27 0.40 0.28 0.50

Gazli 0-75 0-80 1.55 2.00
L. Prieta EW 2.20 4.20 3.20 8.00
El Centro 1.60 240 2.20 3.80
Spitak 1.60 250 2-60 5.50
L. Prieta NS 2.75 650 2-70 600
Mean, p 1.78 3.28 245 506
Std. dev., a 0.67 1-94 055 2.03
cov = alp 0.38 0.59 022 0.40
~ ~~~

Note: Values of ductility < 1.0indicate elastic response.

rnd to o.ng SI.

O - N m b n ~ r - C m ~

Time (sec)
Figure 11. El Centro earthquake time-displacement response, SI and PGA scaling
458 A. S. ELNASHAI AND D. C. McCLURE

o z - y
-
( Y y
(Y

Time (sec)
m y
0
s y m
w

Figure 12. Friuli earthquake timedisplacement response, SI and PGA scaling

150

n
[ 100
v
CI

I"
d o

.g -100
-150
0 2 4 6 8 1 0 1 2 1 4
Time (sec)
Figure 13. Gazli earthquake time-displacement response, SI and PGA scaling

evaluation. More interestingly some important features come to light from the data presented in Tables X
and XI.Although not strictly applicable for such a limited data set, values of the coefficient of variation were
calculated for structural response and response parameters. The outcome suggests less dispersion between
results for scaling to spectra1 intensity, rather than peak ground acceleration.
It is evident that direct acceleration scaling takes no account of levels of spectral response at frequencies
other than that associated with the peak acceleration pulse. Consequently two records which possess equal
peak ground acceleration will be scaled with the same factor irrespective of their frequency content. In
contrast, scaling to spectral intensity ensures that all records possess the same intensity across the response
spectrum band for periods from 01-25 s, and for structures of moderate to long period, constitutes a more
appropriate approach.
The results presented in Table XI indicate that the coefficient of variation of the ductility demand, though
the sample is by no means statistically viable, drops by 30-40 per cent when spectral intensity scaling is used.
 SEISMIC DESIGN PARAMETERS OF RC BRIDGE PIERS 459

spi su*d to 0.2511 S.I.

0 5 10 15 20 25
Time (sec)
Figure 14. Spitak earthquake time-displacementresponse, SI and PGA scaling

.... 4 E W Scaled to 0 . Z g POA

- Lpew scaled to 0.25g SJ.

0 5 10 15 20 25 30
Time (sec)
Figure 15. Loma Prieta EW earthquake time-displacementresponse, SI and PGA scaling

Furthermore, if the behaviour factor is defined in terms of the equivalent design code spectrum values for
a given record,

where the superscript c denotes the equivalent design code spectrum value of acceleration, for a given natural
earthquake record scaled using spectral intensity, it is evident that employing spectral intensity scaling would
produce behaviour factors in significant contrast to those presented in Section 7.3.1, since yield and collapse
ground accelerations will be different from those given in Table VIII.
460 A. S. ELNASHAI AND D. C. McCLURE

LpNS sul#l to 0.Ug S.I.

0 5 10 15 20 25 30
Time (sec)
Figure 16. Lorna Prieta NS earthquake time-displacementresponse, SI and PGA scaling

8. CONCLUSIONS AND RECOMMENDATIONS

This study is aimed at providing guidance regarding the inelastic analysis of reinforced concrete bridges
under earthquake loading. Consequently, model assumptions and input motion variability have been
investigated.
Mesh refinement in static analysis had little effect on the overall load-displacement response. However,
due to relocation of the critical section, local ductility differed by more than 20 per cent. The inclusion of
soil-structure interaction in the static analysis resulted in reduction of 15-30 per cent in displacement
ductility, due to large variations in the yield limit state attainment. Moreover reductions of 25-45 per cent in
displacement ductility ensued when the deck torsional stiffness was introduced. However, the effect of the
variations in the soil lateral stiffness and the deck torsional stiffness are less significant, in comparison with
the effect of including the modelling feature. The plastic hinge length only changed when deck stiffness was
introduced. This is due to the existence of two zones of inelasticity, as opposed to one for the free cantilever.
Furthermore, a 20 per cent increase in displacement ductility was observed.
The investigation into the effect of ground motion characteristics has indicated very large differencesin the
seismic behaviour factor q (R or R , in US. practice). A carefully selected suite of earthquake records,
exhibiting a range of low, medium and high peak ground-acceleration-to-velocityratios, produced widely
varying response parameters. Variations in q between 3.2 and 7.2 were recorded, a difference of 225 per cent.
It is therefore concluded that seismic design codes recommending the use of a minimum of three natural
earthquakes may be inadequate. No correspondence was observed between a/v ratio and behaviour factor.
The most important parameter in seismic demand evaluation using inelastic dynamic analysis is the
method of earthquake scaling. Analysis of the bridge pier under two sets of records, the first scaled to
a common peak ground acceleration, whilst the second scaled to a common velocity spectral intensity
produced significantly different results. The latter suite, on average, imposed ductility demands 54 per cent
higher than those imposed by the former, with some cases in excess of 100 per cent. Moreover, the dispersion
of response parameters amongst the former records was reduced by more than 30 per cent when spectral
intensity scaling was used.
Based on the analysis undertaken above, and constrained by its limitations, the following recommenda-
tions are offered to improve the quality of inelastic dynamic analysis, and to render analyses undertaken by
different research groups more comparable than is currently observed:
-Mesh refinement in the plastic hinge should be adopted. Thus members in the expected zone of plasticity
should be limited to L/lO-L/8, where L is the length of the pier column.
 SEISMIC DESIGN PARAMETERS OF RC BRIDGE PIERS 46 1

- Soil-structure interaction plays a significant role in the response of the system, and thus should be included.
However, only minor variations in response take place across the limits of soil stiffness.
-Displacement ductility is significantly affected by the inclusion of piles. Therefore, it should be used in
seismic assessment only alongside local ductility quantities such as curvature ductility and plastic hinge
length.
-Deck stiffness, if intended to be mobilized, leads to appreciable differences in response parameters, but
actual sensitivity to variation in deck stiffness is minimal. Consequently only an overall idea of the likely
deck dimensions is sufficient to arrive at a stiffness suitable for response parameter calculations.
-In seismic assessment of bridges, a suite of earthquake records spanning the range of peak ground-
acceleration-to-velocityratios should be employed. The minimum number of records within each of the u/u
categories high, medium and low should be two, hence a suite of six earthquake records is required, as
a minimum.
-When calculating seismic demand estimates for bridges, it is essential to use a scaling procedure which has
an influence on the input seismic energy. The velocity spectral intensity, with tl = 0.1 and t 2 = 2.5, is
a feasible technique that leads to results which are significantly more stable than direct acceleration scaling.
Narrowing the period range for spectral intensity scaling may be advantageous, but this has not been
studied herein.

Further work concerning the applicability of the above recommendations to more complicated models of
bridge and soil systems under biaxial and asynchronous ground motion is needed. Various research groups
are currently co-operating on such topics.

ACKNOWLEDGEMENTS
Partial financial support has been provided by the University of Southern California, through the research
programme funded by the Carpenters/Contractors Co-operation Committee Inc. and the Commission of
European Communities (Human Capital and Mobility; Prenormative Research in Support of EC8). The
authors are also grateful for the help of Dr. B. A. Izzuddin in using the program ADAPTIC.

APPENDIX: NOTATION
U acceleration
ag (yield) earthquake record peak acceleration at yield
(yield) design code spectrum peak acceleration at yield
Ug (collapse) earthquake record peak acceleration at collapse
(collapse) design code spectrum peak acceleration at collapse
dCC diameter of the confined concrete core
L O compressive strength of unconfined concrete
fi lateral confining pressure
fs, yield strength of confining stirrups
fc” cylinder compressive strength
f; concrete tensile strength
fY characteristic steel yield
f” ultimate strength of steel
ke deck torsional stiffness
Kcl confinement factor
qcode code behaviour factor
4‘ actual behaviour factor
4* actual behaviour factor calculated using equivalent design code spectrum accelerations,for
a given natural earthquake record, based on spectral intensity scaling
462 A. S. ELNASHAI AND D. C. McCLURE

spacing of the confining stirrups

elastic design spectral acceleration
inelastic design spectral acceleration
elastic design spectral acceleration for collapse
elastic design spectral acceleration for yield
spectral intensity
pseudo-spectral velocity
numerical integration time step
velocity
shear contribution from concrete
shear contribution from steel reinforcement
shear contribution from axial force
pier total shear capacity

Greek letters
a, effective confinement coefficient
B fraction of critical damping
EsY
steel yield strain
Es steel strain
EaY strain at yield
P strain hardening parameter
PA displacement ductility
Pe rotational ductility
P4 curvature ductility
Ps volumetric ratio of confining steel
z normalized plastic hinge length

REFERENCES
1. M. Petrangeli and P. E. Pinto, ‘Seismic response analysis and assessment using nonlinear and linear models’, in Proc. 2nd int.
workshop on seismic design of bridges, Queenstown, New Zealand, 1994, pp. 146-163.
2. G. Monti and C. Nuti, ‘Seismic analysis of a curved continuous bridge’, in Proc. 2nd int. workshop on seismic design of bridges,
Queenstown, New Zealand, 1994, pp. 597-616.
3. J. Kodera, Y. Maehara and J. Watanabe, ‘Analysisof bridge Example A modal analysis under two different supporting conditions’,
in Proc. 2nd int. workshop on seismic design of bridges, Queenstown, New Zealand, 1994, pp. 665-684.
4. M. J. Kowalsky and M. J. N. Priestley, ‘Substitutestructure analysis of modern curved bridge’, in Proc. 2nd int. workshop on seismic
design of bridges, Queenstown, New Zealand, 1994, pp. 203-221.
5. S . P. Singh and G. L. Fenves, ‘Earthquake response of structure a using nonlinear dynamic analysis’, in Proc. 2nd int. workshop on
seismic design of bridges, Queenstown, New Zealand, 1994, pp. 222-239.
6. S . Dodd, A. S. Elnashai, B. A. Izzuddin and G. M. Calvi, ‘A 3D nonlinear dynamic analysis of a curved bridge’, in Proc. 2nd int.
workshop on seismic design of bridges, Queenstown, New Zealand, 1994, pp. 617-640.
7 . G. A. Chan and J. B. Mander, ‘Seismic energy based fatigue damage analysis of bridge columns: part I1 - evaluation of seismic
demand‘, National Centre for Earthquake Engineering Research Technical Report NCEER-94-0013, 1994.
8. P. J. Madas and A. S . Elnashai, ‘A theoretical model for composite beam-columns under cyclic loading’, Imperial College,
Engineering Seismology and Earthquake Engineering Report No. 89/10, 1989.
9. EC8, EurocodeS: Earthquake Resistant Design of Structures, ENV1998, CEN, 1992.
10. J. B. Mander, M. J. N. Priestley and R. Park, ‘Theoretical stress-strain model for confined concrete’, J . struct. eng. ASCE 114,
1804-1826 (1988).
11. A. S. Elnashai and J. G. Beith, ‘Seismic design limit states and analytical ductility supply of circular bridge piers’, in Proc 2nd int.
workshop on seismic design of bridges, Queenstown, New Zealand, 1994, pp. 310-364.
12. G. Lappas and T. P. Tassios, ‘Estimation of behaviour factors of RC buidlings’, Eur. earthquake eng. 3, 38-42 (1988).
13. R. Park (ed.), ‘Comparative bridge examples’, in Proc 2nd int. workshop on seismic design of bridges, Queenstown, New Zealand,
1994, pp. 567-578.
14. T. J. Zhu, A. C. Heidebrecht and W. K. Tso, ‘Effect of peak ground acceleration to velocity ratio on ductility demand of inelastic
systems’, Earthquake eng. struct. dyn. 16, 63-79 (1988).
15. NBCC Associate Committee on the National Building Code, National Building Code of Canada 1985, National Research Council of
Canada, Ottawa, Ontario, 1985.
16. B. M. Broderick, ‘Seismic testing, analysis and design of composite frames’, Ph.D. Thesis, Imperial College, University of London,
1994.
 SEISMIC DESIGN PARAMETERS OF RC BRIDGE PIERS 463

17. A. J. Kappos, ‘Analytical prediction of the collapse earthquake for RC buildings: suggested methodology’, Earthquake eng. struct.
dyn. 20,167-176 (1991).
18. J. A. Blume, N. M.Newmark and L. H. Corning, Design of Multistory Reinforced Concrete Buildings for Earthquake Motions,
Portland Cement Association, 1961.
19. W. K.Tso,T.J. Zhu and A. C. Heidebrect, ‘Engineering implications of ground motion A/V ratio’, Soil dyn earthquake eng. 11,
133-144 (1992).
20. A. M.Chandler, ‘Evaluation of site-dependent spectra for earthquake-resistantdesign of structures in Europe and North America’,
Proc. ICE, London, Part 1,90,60-626 (1991).
21. G. W. Housner, ‘Spectrumintensity of strong-motionearthquakes’, in Proc. symp. earthquake and blast fleets on structures, UCLA,
1952, pp. 20-36.
22. J. M.Nau and W. J. Hall, ‘Scaling methods for earthquake response spectra’, J. struct. eng. ASCE 110, 1533-1548 (1984).