Effect of Frame-Restoring Force Characteristics on the Pounding Response of Multiple-Frame Bridges

Susendar Muthukumar,a M.EERI, and Reginald DesRoches,b M.EERI

This study examines the effect of column hysteretic behavior on the impact response of adjacent frames in multiple-frame bridges. A simplied planar analytical bridge model is developed including inelastic frame action, nonlinear hinge behavior, and abutment effects. Pounding is simulated using a stereomechanical approach. The frame hysteretic models considered include the elasto-plastic and bilinear traditional , Q-Hyst stiffness-degrading , and pivot hysteresis strength-degrading models. Analytical studies conducted on adjacent bridge frames reveal that the traditional models underestimate the stiff frame displacement amplication due to pounding, and overestimate the exible frame displacement amplication, when compared with other hysteretic models. A stiffness-degrading model is recommended to accurately estimate the pounding response of bridge frames subjected to far-eld ground motion. The use of a strength-degrading model increases the stiff frame displacement amplication by 125% when compared to the stiffnessdegrading model for highly out-of-phase frames, and is recommended in the presence of near-eld ground motions. DOI: 10.1193/1.2103107 INTRODUCTION Bridges are the lifeline of a highway transportation network and past earthquakes have illustrated that they are vulnerable to severe damage and/or collapse during moderate to strong ground motion. Among the possible structural damages, seismic-induced pounding has commonly been observed in several recent earthquakes. The 1994 Northridge earthquake revealed substantial impact damage at the expansion hinges and abutments of standing portions of the connectors at the Interstate 5/State Road 14 interchange that were located at close proximity to the epicenter EERI 1995a . Reconnaissance reports from the 1995 Kobe earthquake identify pounding as a major cause of fracture of the bearing supports and potential contributor to the collapse of the bridge decks EERI 1995b . Hammering at the expansion joints in some bridges resulted in damage to shear keys, bearings, and anchor bolts during the 1999 Chi-Chi earthquake in Taiwan EERI 2001a . Cracking and spalling at expansion joints of concrete bridges were observed during the 2001 Nisqually, Washington, earthquake EERI 2001b . More recently, pounding of adjacent simply supported spans resulting in failure of girder ends and bearing damage was observed during the 2001 Bhuj earthquake in Gujarat, India EERI 2002 .
a b

Browder + LeGuizamon & Associates, 174 West Wieuca Road NE, Atlanta, GA 30342-3220 School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, GA 30332-0355

1113

Earthquake Spectra, Volume 21, No. 4, pages 11131135, November 2005; 2005, Earthquake Engineering Research Institute

1114

S. MUTHUKUMAR AND R. DESROCHES

Figure 1. Types of bridges vulnerable to seismic pounding: a multiple-frame bridge, and b multi-span simply supported bridge.

The multiple-frame bridge and the multi-span simply supported bridge shown in Figure 1 are most susceptible to pounding damage due to numerous independent components and lack of continuity in the structure. In a multiple-frame bridge, the interaction between adjacent frames can result in pounding at the intermediate hinge locations or at the abutments. Pounding of girder ends at the pier locations and end abutments can occur in a multi-span simply supported bridge. Based on observations from past earthquakes, seismic pounding can lead to local crushing and spalling of concrete, result in damage to column bents, abutments, shear keys, bearing pads, and restrainers, and possibly contribute to the collapse of deck spans. Past research into seismic pounding has focused primarily on determining the factors affecting pounding DesRoches and Muthukumar 2002 , modeling the impact phenomenon Jankowski et al. 1998, Malhotra 1998 , studying the effects of pounding on the bridge response Kim and Shinozuka 2003 , and developing mitigation strategies for pounding hazard reduction Kawashima and Shoji 2000, Shinozuka et al. 2000 . Typically, the participating structural systems have been modeled using bilinear or stiffnessdegrading models. However, experimental tests on concrete columns have shown that strength degradation occurs under increased cycles of loading, which is accelerated under the presence of axial compressive loads Saatcioglu and Ozcebe 1989 . To the authors knowledge, no study has yet considered the effects of strength-degrading columns on the pounding response of bridges. This paper investigates the inuence of frame hysteretic characteristics, such as stiffness degradation, strength deterioration, and pinching on the pounding response of a typical multiple-frame bridge. Several analytical models are considered for the frame hysteretic behavior, including the bilinear, Q-Hyst stiffness-degrading , and pivot hysteresis strength-degrading models. A simplied planar nonlinear analytical model of a multiple-frame bridge is developed. Inelastic frame action, nonlinear hinge behavior, and the effects of abutments are modeled. Pounding is simulated using the stereome-

EFFECT OF FRAME-RESTORING FORCE CHARACTERISTICS ON THE POUNDING RESPONSE

1115

Figure 2. Lateral load-deection relation for a reinforced concrete column obtained through experiment Saatcioglu and Ozcebe 1989 .

chanical approach. Effects of skew, curvilinear bridge geometry, and multi-support excitation are not considered. Two adjacent frames of the bridge are isolated and parameter studies are conducted using a suite of far-eld ground motion records to evaluate the effect of various hysteretic models on the frame pounding responses. The pounding response of the various hysteretic models in the presence of near-eld earthquakes is also assessed. Finally, the differences in the global responses of a multiple-frame bridge with restrainers, bearings, and abutments when various hysteretic models are used, are highlighted through an example. HYSTERETIC MODELS FOR REINFORCED CONCRETE BRIDGE COLUMNS Reinforced concrete bridge columns can develop inelastic deformations and exhibit nonlinear behavior under moderate to strong base excitation. In the past, elasto-plastic and bilinear models were used due to their simplicity in concept and numerical implementation. Stiffness degradation in concrete was rst accounted for with the introduction of a degrading stiffness approach Clough and Johnston 1966 . Subsequent experimental tests on both small-scale and full-scale column specimens have shown that cyclic behavior of reinforced concrete is characterized by constantly changing stiffness, strength degradation, and a reduction in energy absorption capacity Takeda et al. 1970, Saatcioglu and Ozcebe 1989, Dowell et al. 1998 . A typical lateral load-deection hysteretic relationship for a reinforced concrete column is shown in Figure 2. The general hysteretic characteristics can be summarized as follows:

Reduction in stiffness occurs with increased displacements, which can be attributed to the exural cracking in concrete and the Bauschinger effect in steel.

1116

S. MUTHUKUMAR AND R. DESROCHES

Figure 3. a Elasto-plastic model; b bilinear model.

The peak strength attained in each cycle decreases with increased loading cycles.
This strength degradation is a result of the disintegration of core concrete. The hysteretic loop exhibits pronounced pinching effects, which can be attributed to high shear stress reversals and slippage of the longitudinal reinforcement within the anchorage area. The hysteretic characteristics of reinforced concrete are dependent on the loading history.

Several hysteretic models have been developed to capture the nonlinear dynamic response of reinforced concrete columns subjected to base excitation. These range from relatively simplistic models such as the elasto-plastic and bilinear models, to more rigorous models such as the Takeda Takeda et al. 1970 , Park Kunnath et al. 1990 , and pivot hysteresis models Dowell et al. 1998 . Other models such as the Clough model Clough and Johnston 1966 and the Q-Hyst model Saiidi and Sozen 1979 have also been popular. A brief discussion of the hysteretic models considered in this study is provided below.
ELASTO-PLASTIC MODEL

This is a simple model dened by three rules. The backbone curve is dened by an elastic stiffness k that represents cracked-section behavior and a post-yield portion with zero stiffness, as shown in Figure 3a. The unloading stiffness is taken to be the same as the elastic loading stiffness. This model is a very poor representation of the hysteretic behavior of concrete as it does not represent stiffness deterioration with increasing displacement amplitude reversals. However, it has been extensively used because of its simplicity in modeling.
BILINEAR MODEL

This is very similar to the elasto-plastic model, but it also accounts for the strain hardening effect in steel using non-zero post-yield stiffness, as shown in Figure 3b. Stiffness and strength degradation effects cannot be represented. Both the elasto-plastic and

EFFECT OF FRAME-RESTORING FORCE CHARACTERISTICS ON THE POUNDING RESPONSE

1117

Figure 4. Q-Hyst model for reinforced concrete.

bilinear models do not consider hysteretic energy dissipation for small displacements. Many studies evaluating the effects of pounding have used bilinear models to represent the behavior of adjacent structures Anagnostopoulos 1988, Pantelides and Ma 1998, Kim et al. 2000 .
Q-HYST MODEL

The Q-hyst model Saiidi and Sozen 1979 is dened by four rules and closely represents the response from a Takeda model, which is a more realistic representation of the cyclic behavior of reinforced concrete columns. The backbone curve used is bilinear with strain hardening as shown in Figure 4. Stiffness degradation is accounted for at unloading and load reversal. The unloading stiffness is dened by Kq = K Dy / D 0.5, where K is the initial elastic slope, D is the largest absolute deformation, and Dy is the yield deformation. The reloading stiffness Kp is dened as the slope of the line connecting the intersection of the latest unloading branch with the displacement axis point A to the maximum absolute displacement point B , as shown in Figure 4. The Q-Hyst model is much simpler than the Takeda model. Neither the Q-Hyst and Takeda models, however, account for the effect of column axial loads and strength degradation in concrete.
PIVOT HYSTERESIS MODEL

The pivot model Dowell et al. 1998 is governed by three simple rules and has the ability to capture the dominant nonlinear characteristics of concrete under cyclic load. The backbone curve used for positive and negative loading is shown in Figure 5. The rst and second branches of the strength envelope represent cracked-section stiffness and strain-hardening stiffness, respectively. Strength degradation from shear failure or connement failure is represented by the third branch. The fourth branch allows for a linearly decreasing residual strength.

1118

S. MUTHUKUMAR AND R. DESROCHES

Figure 5. Strength envelope for the pivot hysteresis model.

Primary pivot points P1 through P4 control the amount of softening expected with increasing displacement, using parameters 1, 2 as shown in Figure 5. Pinching pivot points PP2 and PP4 x the degree of pinching following a load reversal, through parameters *, *. The response follows the strength envelope as long as no displacement re1 2 versal occurs. Once the yield displacement is exceeded in either direction, a modied strength envelope is dened by the lines joining PP4 to S1 and PP2 to S2, as illustrated in Figure 5. The pinching pivot points start moving toward the origin of the forcedeformation relation, once strength degradation occurs. The pinching parameters, *, 1 * 2, are given by the following equations:
* i

i ;dimax

d ti d ti

1 2

* i

Fimax F ti

i ;dimax

where 1, 2 dene the degree of pinching for a ductile exural response before strength degradation occurs. Displacements dimax and dti represent the maximum displacement

EFFECT OF FRAME-RESTORING FORCE CHARACTERISTICS ON THE POUNDING RESPONSE

1119

Figure 6. Analytical model of multiple-frame bridge used in study.

and strength degradation displacement, respectively, in the ith direction of loading. Fimax and Fti represent the force levels corresponding to dimax and dti, respectively. The primary advantage of the pivot hysteresis model when compared to the other models is its ability to represent effects of cyclic axial load due to frame action, unsymmetrical sections, and strength degradation. Cyclic actual load effects are achieved with predened parameters, and the more general variation in axial load due to vertical accelerations is not included. Unlike other models, the pivot model recognizes that yielding in one direction does not soften the member in the opposite loading direction. For instance, if the yield strength is exceeded in the positive loading direction, unloading occurs and the member reloads in quadrant Q2 towards PP2, the response will follow the initial elastic loading line if yielding has not yet occurred in the negative loading direction. This study assumes symmetric sections for the bridge columns and only the effects of strength degradation will be considered. ANALYTICAL BRIDGE MODEL A simplied planar nonlinear analytical model of an n-frame bridge is developed, as shown in Figure 6. Each bridge frame is idealized as a single-degree-of-freedom SDOF yielding element with mass mi, initial stiffness ki, and a viscous damping coefcient ci. For the frame force-deformation relation, any of the hysteretic models described in the previous section can be used. The elastomeric bearings at the hinge locations and abutments are modeled using bilinear spring elements, following Kellys model Naeim and Kelly 1999 . Cable restrainers at the intermediate hinges are modeled using tension-only bilinear elements 5% strain hardening with a slack. Abutments are modeled using linearized springs with different properties in active and passive action. The equations of motion for the bridge system subjected to horizontal ground motion are

1120

S. MUTHUKUMAR AND R. DESROCHES

m1 . . mn

u1 . . un

c1 + . . cn

u1 . . un

F F1 u 1 + . . F Fn u n

F R1 u 2 u 1 F R2 u 3 u 2 F R1 u 2 u 1 FRn1 . un un1

F B1 u 2 u 1 F B0 u 1 u 0 F B2 u 3 u 2 F B1 u 2 u 1 F Bn m1 = . . mn = F B0 u 1 u 0 FBn un+1 un . un+1 un FBn1 un un1 1 . ug . 1

F I1 u 2 u 1 + F I2 u 3 u 2 F I1 u 2 u 1 FIn1 . un un1 F A1 u 0 FA2 un+1

3a

3b

where FFi is the inelastic restoring force for each frame based on the hysteretic relation chosen, FRi is the force from restrainer Ri, FBi is the force in bearing Bi, FIi is the force due to impact between frames i and i + 1, and FAi is the force in abutment Ai; ui, ui, and ui i = 1 to n represent the frame acceleration, velocity, and displacement relative to the ground; u0, un+1 are abutment displacements obtained using static condensation; and ug represents the horizontal ground motion applied to the bridge. The solution of Equation 3 is obtained numerically using the fourth-order Runge-Kutta method Kreyzig 1999 . Seismic pounding occurs when the relative displacement between adjacent frames ui ui+1 exceeds the hinge gap gp . Pounding is accounted for using the stereomechanical approach Goldsmith 1960 , which assumes that impact is instantaneous. The momentum balance principle is used along with the coefcient of restitution to model impact. The force due to impact is taken as zero. However, the velocities of the colliding masses are modied at the instant of impact as

v1 = v1 1 + e v2 = v2 + 1 + e

m2 v1 v2 m1 + m2 m1 v1 v2 m1 + m2

4a

4b

where vi and vi+1 are the velocities of adjacent frames after impact; vi, vi+1 are the frame velocities before impact; and e is the coefcient of restitution, which is assumed as 0.8 in this study.

EFFECT OF FRAME-RESTORING FORCE CHARACTERISTICS ON THE POUNDING RESPONSE

1121

Figure 7. Single-degree-of-freedom system used to compare hysteretic model responses.

Other approaches to model pounding include the contact force-based linear spring element Maison and Kasai 1992 , Hertz nonlinear spring Davis 1992 , and the Kelvin element Jankowski et al. 1998 . The high impact stiffness used in the contact forcebased approaches produces acceleration spikes at the instances of impact, which appear in the dynamic response. The stereomechanical approach does not produce these acceleration spikes. Furthermore, the stereomechanical method is momentum-based, and thus the uncertainty associated with the value of impact spring stiffness is eliminated. The stereomechanical approach is no longer valid, however, if the impact duration is large enough so that signicant changes occur in the conguration of the colliding bodies. Recently, a study has been conducted to address the validity of the various approaches to model pounding Muthukumar and DesRoches 2004 . The results indicate that the stereomechanical and contact force-based approaches produce similar displacement responses for a given coefcient of restitution. The accelerations are much smaller for the stereomechanical approach. Abutments play an important role in the seismic response of a bridge, as they attract a large portion of the earthquake loads and many design guidelines require their inclusion as equivalent linear springs. Impact between the bridge deck and end abutments will induce high passive pressures, and hence the effect of deck impact is indirectly considered by including a high passive stiffness in the abutment model. The passive stiffness of the abutment is determined using the Caltrans procedure Caltrans 1999 . The stiffness is calculated to be around 2600 kips/ in for a backwall height of eight feet and an acceptable deformation of one inch. COMPARISON OF HYSTERETIC MODEL RESPONSES An SDOF system shown in Figure 7 is considered to study the differences in frame response when various hysteretic models are used. The system has an initial stiffness,

1122

S. MUTHUKUMAR AND R. DESROCHES

Figure 8. Unloading stiffnesses: a Q-Hyst model, and b pivot model.

K = 295 kips/ in, damping ratio, = 5%, and a period, T = 1 second. The Saratoga-Aloha Avenue record with a peak ground acceleration PGA of 0.51 g, from the 1989 Loma Prieta earthquake is chosen for analysis. The yield strength of the system is selected such that the target ductility is = 4 when the Q-Hyst model is used as the frame forcedeformation relation. Five percent strain hardening is used wherever applicable.
The hysteretic parameters of the Q-Hyst model unloading stiffness KqQH, reloading stiffness KpQH and the pivot model without strength degradation pivot parameters 1 , 2, and the pinching pivot parameters 1 , 2 need to be correlated so that their responses are as close to each other as possible. Consider the unloading stiffnesses of the two models, as illustrated in Figure 8. From Figure 8a, KqQH can be expressed as

KqQH = K

Dy K = Dm

where is the ductility ratio from the Q-Hyst model. From Figure 8b, the unloading stiffness for the pivot model, KqPH can be written as

EFFECT OF FRAME-RESTORING FORCE CHARACTERISTICS ON THE POUNDING RESPONSE

1123

Figure 9. Reloading stiffnesses: a Q-Hyst model, and b pivot model.

KqPH =

K 1

1+

where 1 = 2 = pivot parameter . Equating Equations 5 and 6, an expression for the pivot parameter can be found in terms of the strain hardening ratio * and the ductility ratio of the Q-Hyst model, as given below.

= 1

1+

Figure 9 sketches the reloading stiffnesses for the two models. For the Q-Hyst model, the reloading stiffness KpQH can be written as

1 1 * 1 * 2 * Assuming 1 = 2 = , the pinching pivot parameter can be expressed as KpQH = K = KpPHXr Kp Fy 1 K

where Xr is x-coordinate of point A in Figure 9b and can be expressed as

Xr =

Fy 1 K

10

where Fy is the yield strength of the system and K is the initial elastic stiffness. Setting KpQH = KpPH and using Equations 8 and 10, the pinching pivot parameter can be simplied to

1124

S. MUTHUKUMAR AND R. DESROCHES

Figure 10. Responses of SDOF system with various hysteretic models subjected to the 1989 Saratoga-Aloha Avenue record: a time history of displacements, and b hysteresis loops.
*

1 =

+ 2 * +

* 3/2

1
*

11

1+ +

1
*

It should be noted that the pivot model assumes that yielding in one direction does not soften the member in the opposite direction. Thus the response proceeds towards point C from point PP4, if the yield deformation has not been exceeded in the negative loading direction, in Figure 9b. On the other hand, the response from the Q-Hyst model will proceed toward point B, the largest absolute displacement. This implies that in most cases, the maximum response from the pivot model will either be equal to or smaller than the maximum Q-Hyst model response. For a target ductility, = 4, and a strain hardening ratio, * = 5%, parameters and can be determined as 1.70 and 0.43, respectively. The responses from the various hysteretic models when the SDOF system is subjected to the Saratoga-Aloha Avenue record are presented in Figure 10. The time-history responses from all the models are identical for the rst 6.2 sec, as nonlinear deformations have not yet occurred. However, once the yield force has been exceeded, the elasto-plastic and bilinear model responses exhibit more permanent deformations, with a pronounced shift in the equilibrium position. This is because neither of the two models considers hysteretic energy dissipation for small displacements. Thus, should a load reversal occur immediately after the reloading stage, there will be no energy dissipation from the elasto-plastic and bilinear models. The Q-Hyst and pivot models will have some energy dissipation, however, since their unloading and reloading stiffnesses are generally different. The absolute maximum displacement from the Q-Hyst model is larger than the corresponding bilinear response, as stiffness degradation in the Q-Hyst model produces less damping per cycle. However, despite major differences in the force-deformation rela-

EFFECT OF FRAME-RESTORING FORCE CHARACTERISTICS ON THE POUNDING RESPONSE

1125

Figure 11. Two-degree-of-freedom model of adjacent frames used for hysteretic model parameter study. Fundamental periods, T1 = 2 m1 / k1; T2 = 2 m2 / k2.

tions, the absolute peak displacements from the Q-Hyst and elasto-plastic models are identical 6.3 inches . Typically, the Q-Hyst response is expected to be larger than the elasto-plastic response due to a smaller hysteretic loop for the Q-Hyst model. But for this particular ground motion, the elasto-plastic model shows large excursions along the post-yielding branch, which could account for the peak displacements being identical. The lack of strain hardening in the elasto-plastic model could be a factor as well. The hysteretic loops from the pivot without strength degradation and Q-Hyst models are very similar, with only a 15% difference in the maximum displacement response, for this particular ground motion record. The disparity can be attributed to the assumption in the pivot model that yielding in one direction does not soften the member in the opposite loading direction. Preliminary studies indicate that the maximum displacements from both the models can also be identical depending on the ground motion record. Thus, for the purposes of comparing the maximum displacement response, the correlation of hysteretic parameters between the Q-Hyst and the pivot models given by Equations 7 and 11 appears to be satisfactory and will be utilized through the rest of the study. PARAMETER STUDY TO COMPARE THE IMPACT RESPONSE OF VARIOUS HYSTERETIC MODELS To investigate the inuence of column hysteretic characteristics on the pounding response of multiple-frame bridges, two adjacent frames are isolated and a parameter study is conducted with the two-degree-of-freedom idealized system shown in Figure 11. The structure on the left represented by a thick line is taken as the stiff frame and the structure on the right is assumed to be the exible frame. The fundamental periods of

1126

S. MUTHUKUMAR AND R. DESROCHES

the stiff and exible frames are T1, and T2, respectively. Restrainers, bearings, and abutments are not included so that differences in frame pounding responses can be directly attributed to differences in hysteretic model characteristics. A previous study by the authors identied the frame period ratio T1 / T2 as the primary factor affecting the seismic pounding response DesRoches and Muthukumar 2002 . Smaller period ratios T1 / T2 0.3 represent highly out-of-phase frames and period ratios closer to one T1 / T2 0.7 indicate that the frames are more in-phase. Based on a normalization of the governing equations of motion, the ground motion effective period ratio T2eff / Tg = T2 / Tg was identied as another parameter affecting the pounding response. The characteristic period of ground motion Tg is dened as the period at which the input energy of a 5% damped linear elastic system is a maximum. The results of the study showed that the frame displacement amplication due to pounding can be classied into three zones depending on the ground motion effective period ratio, T2eff / Tg. Impact was found to be most detrimental in Zone I T2eff / Tg 1 . Thus, in the following study, only Zone I responses from the various hysteretic models are considered, with three values for the frame period ratio: T1 / T2 = 0.3 highly out-of-phase frames , T1 / T2 = 0.5 moderately out-of-phase frames , and T1 / T2 = 0.7 essentially inphase frames . The exible frame period is xed at 0.40 second and the stiff frame period is varied to get the desired period ratio. The yield strength of the system is selected such that the target ductility is = 4 when the Q-Hyst model is used. Five percent strain hardening is assumed, wherever applicable. The effect of pounding is expressed in terms of frame displacement amplication , which is the ratio of the maximum pounding displacement to the maximum frame displacement if pounding does not occur. The hinge gap is 1 set very large for the no-pounding analysis, and is assumed as 2 inch for the pounding analysis. The parameters of the pivot model are selected such that strength degradation does not occur during the no-pounding analysis. The following values are used for the various parameters: pivot parameters, 1 , 2 = 1.70, pinching pivot parameters, 1 , 2 = 0.43, strength degradation ductility, + , = 4.0, residual strength ratio, Fdr+ , Fdr = 0.7, ret t sidual strength reduction ductility, + , = 8, failure ductility, + , = 100. The corred d f f lation of the and parameters with the Q-Hyst model parameters ensures that the ductility of each frame when the pivot model is used and when no pounding occurs is close to 4. Thus differences in displacement amplications between the Q-Hyst and pivot models can be directly related to the effects of strength degradation. Ten far-eld ground motions recorded on medium soil Tg = 0.6 1.2 sec are selected for analysis from the PEER Strong Motion Database http://peer.berkeley.edu/ smcat/ . The characteristic motions include the Pasadena record from the 1971 San Fernando earthquake, the Waho record from the 1989 Loma Prieta earthquake, and the Wonderland Avenue record from the 1994 Northridge earthquake. The pseudo acceleration response spectra and record names are presented in Figure 12a. Each record is scaled such that the spectral acceleration at fundamental period equals the mean spectral acceleration of the suite of records at the fundamental period of the system T

EFFECT OF FRAME-RESTORING FORCE CHARACTERISTICS ON THE POUNDING RESPONSE

1127

Figure 12. Pseudo acceleration spectra of records used in analysis: a far eld, and b near eld.

= 0.40 sec . The yield strengths of the system for each ground motion record at the various frame periods are determined using an iterative scheme. The characteristic periods of the records Tg ensure that the ground motion effective period ratio, T2eff / Tg, lies in Zone I.
Figure 13 presents the mean plus one standard deviation of the displacement amplication due to pounding for the various hysteretic models as a function of the frame period ratio T1 / T2 , for ground motion effective period ratios in Zone I T2eff / Tg 1 . In general, the elasto-plastic and bilinear models traditional models underestimate the stiff frame amplication and overestimate the exible frame amplication, when compared to the Q-Hyst and pivot models advanced models . For instance, at T1 / T2 = 0.3, the stiff frame displacement amplication predicted by the advanced models is 30% more than that predicted by the traditional models. The traditional models underestimate

Figure 13. Mean plus one standard deviation of displacement amplication due to pounding from various hysteresis models10 far-eld ground motion records; Zone I T2eff / Tg 1 .

1128

S. MUTHUKUMAR AND R. DESROCHES

the exible frame displacement amplication by 20%, when T1 / T2 = 0.3. The differences become smaller with increasing period ratio. At T1 / T2 = 0.5, the differences between the traditional and advanced model responses are 20% for the stiff frame and 10% for the exible frame. For essentially in-phase frames T1 / T2 = 0.7 , the deviations are only 5% and 2% for the stiff and exible frames, respectively. While comparing the Q-Hyst and pivot models, the strength degradation effect imposes no additional demands on the response of the system. The pivot model only shows a 7% increase in the mean displacement amplication when compared to the stiffnessdegrading Q-Hyst model, for T1 / T2 = 0.3. In fact, at T1 / T2 = 0.5, the stiff frame amplication from the pivot model is smaller than the Q-Hyst model response by approximately 12%. All of the hysteretic models predict similar displacement amplications when the frames are essentially in-phase T1 / T2 = 0.7 . The coefcient of variation COV , dened as the ratio of the standard deviation to the mean, ranges from 55% at low-period ratios T1 / T2 = 0.3 to 14% at high-period ratios, for the pivot hysteresis model. The COVs for the Q-Hyst model range from 34% to 14%, for low to high frame period ratios. The results indicate that the effects of pounding are highly dependent on the frame hysteretic model, especially for out-of-phase frames. The selection of traditional models like elasto-plastic and bilinear models can result in an underestimation of the impact amplications for the stiff frame and predict higher amplications for the exible frame when compared with more advanced models. The effects of strength degradation in predicting the pounding response of adjacent frames are not signicant as long as stiffnessdegradation is modeled. A parameter study is presented in the following subsection, to study the effects of strength degradation and pounding in the presence of near-eld ground motions.
EFFECT OF NEAR-SOURCE GROUND MOTIONS

Near-eld earthquake motions are characterized by high peak ground accelerations and velocity pulses with a long-period component Yang and Agrawal 2002 . Such characteristics may greatly amplify the dynamic response of multiple-frame bridges, resulting in severe damage. Such recent earthquakes as the 1994 Northridge, 1995 Kobe, 1999 Kocaeli, and 1999 Chi-Chi earthquakes have demonstrated the damage that can be caused by near-eld ground motions. To study the effects of near-eld ground motion on the pounding response of strength-degrading bridge frames, ten near-source records Tg = 0.6 1.2 sec are selected for analysis from the PEER Strong Motion Database http://peer.berkeley.edu/smcat/ . The characteristic ground motions considered include the Bonds Corner record from the 1979 Imperial Valley earthquake, the Sylmar record from the 1994 Northridge earthquake, and the TCU129N record from the 1999 Chi-Chi earthquake in Taiwan. The two-degree-of-freedom system considered in the earlier section is adopted for analysis. The frame yield strengths at various periods for each ground motion record are obtained such that the frame ductility demands equal four when the Q-Hyst model is used. All records are scaled to the mean spectral acceleration at the

EFFECT OF FRAME-RESTORING FORCE CHARACTERISTICS ON THE POUNDING RESPONSE

1129

Figure 14. Mean plus one standard deviation of displacement amplication due to pounding from various hysteresis models10 near-eld ground motion records; Zone I T2eff / Tg 1 .

fundamental period of the system. Figure 12b presents the pseudo acceleration response spectra and the record names for the near-eld ground motions. The mean spectral acceleration at the fundamental period T = 0.40s is 0.83 g. The mean plus one standard deviation of the displacement amplication due to pounding for the various hysteretic models is presented in Figure 14. As observed for the far-eld records, the traditional models elasto-plastic and bilinear models underestimate the stiff frame amplication and overestimate the exible frame amplication when compared to the more advanced models Q-Hyst and pivot models . However, when using near-source records, the differences between the traditional and advanced models persist even when the frames get more in-phase T1 / T2 = 0.5, 0.7 , unlike the case with far-eld ground motions. At T1 / T2 = 0.7, the traditional models underestimate the stiff frame displacement amplication by 20% and overestimate the exible frame amplication by 15% when compared to the more rigorous models. The biggest difference in using near-eld records is that strength degradation and pounding signicantly affect the frame response, especially when the frame is highly out-of-phase. The pivot model results in a mean stiff frame displacement amplication of 5.8 as opposed to 2.6 for the Q-Hyst model. Large amplications are observed for several ground motions, including the 1984 Halls Valley, 1994 Sylmar, and 1999 TCU076N records. Thus accounting for strength degradation increases the stiff frame demand by 125%. The corresponding increase for the case with far-eld records is only 7%. In the presence of near-eld ground motions, a bilinear or stiffness-degrading model for the column will grossly underestimate the pounding demands when compared with a strength-degrading model, for highly out-of-phase frames. The earlier study using far-eld ground motions indicated that the frame amplications get closer to unity as the period ratio becomes higher. The pivot model showed a stiff frame amplication of 1.07 and a exible frame amplication of 0.95, at T1 / T2

1130

S. MUTHUKUMAR AND R. DESROCHES

Table 1. Properties of various bridge components used in study

Element Frame Component Initial stiffness kips/in 1333 577 Initial stiffness kips/in 200 100 6 2600 Active stiffness kips/in 10 Yield strength kips 877 750 Yield strength kips 840 420 2.4 1560 Period s 0.47 1.12 Strain hardening % 5 5 33 33

F1 , F4 F2 , F3

Element Restrainer Bearing

Component

R1 , R3 R2 B1 , B2 , B3 B0 , B4

Element Abutment

Component

Passive stiffness kips/in 2600

A1 , A2

= 0.7. However, for near-eld ground motions, the frame amplications show greater discrepancy from unity. In the latter case, the corresponding pivot model amplications are 1.3 and 0.85 for the stiff and exible frames, respectively.
EFFECTS OF HYSTERETIC MODEL ON THE GLOBAL POUNDING RESPONSE In this section, the differences in the global responses of a multiple-frame bridge system, due to various hysteretic frame models are investigated. The bridge considered consists of four frames connected at three intermediate hinges. The hinge gap is taken as 1 2 in. at all intermediate hinge locations. The simplied bridge model, as shown in Figure 6 is developed with frame weights of 2880 k, 7080 k, 7080 k, and 2880 k, for frames 1 through 4, respectively. The damping ratio for each frame is taken as 5%. The properties of various elements used in the model are listed in Table 1. The hysteretic models discussed earlier, namely, the elasto-plastic, bilinear, Q-Hyst, and pivot hysteresis models are used to describe the frame behavior, with all models having the same initial stiffness and yield strength. The bilinear and Q-Hyst models assume a strain hardening ratio of 5%. The pivot hysteresis model is assumed to have the same properties in both loading directions with dt = 2*dy, dd = 4*dy, and df = 6*dy. The restrainers are designed according to the procedure recommended by DesRo1 ches and Fenves DesRoches and Fenves 2001 . The restrainer slack is assumed as 2 in. The properties for the elastomeric bearings at the hinge locations are calculated based on the bearing dimensions 12 in. 8 in. 4 in., L W H . The bearings at the abutment locations are designed to have a stiffness proportional to the passive stiffness of the abut-

EFFECT OF FRAME-RESTORING FORCE CHARACTERISTICS ON THE POUNDING RESPONSE

1131

Figure 15. Time history of displacementstiff frame frame 1 ; 1989 Saratoga record PGA = 0.5g .

ment. Since the abutment and the bearing at the abutment are springs in series, the active stiffness of the abutment is taken proportional to the stiffness of the hinge bearing. The bridge is subjected to horizontal ground motion from the 1989 Loma Prieta earthquake. The Saratoga record is used, which has a peak ground acceleration of 0.5 g, and a characteristic period Tg of 1.8 second. To study the effect of pounding on the bridge response, two cases are considered; Case 1, where the hinge gap is set very large so that 1 pounding does not occur, and Case 2, where the hinge gap is set at 2 inch and pounding occurs. Figure 15 presents the displacement time history of the stiff frame frame 1 for the various hysteretic models. The corresponding hysteresis loops for the pounding and no-pounding cases are shown in Figure 16. The no-pounding responses for the various models are very similar because there are not too many excursions into the nonlinear range and the displacement ductility is small 2 . However, for Case 2, seismic pounding amplies the displacement response of frame 1 by 100% to 173%, depending on the hysteretic model. The maximum displacement from the Q-Hyst and bilinear models is around 2.0 in, for the pounding case, while the pivot hysteresis model response is 3.0 in.

1132

S. MUTHUKUMAR AND R. DESROCHES

Figure 16. Hysteresis loops for stiff frame frame 1 1989 Saratoga record PGA= 0.5g .

EFFECT OF FRAME-RESTORING FORCE CHARACTERISTICS ON THE POUNDING RESPONSE

1133

The results indicate that strength degradation in bridge columns has a signicant inuence on the pounding response of the stiff frame frame 1 . Strength degradation with increased loading cycles combined with the interaction of adjacent frames increases the stiff frame displacement demand by 50% when compared to other hysteretic models. This example serves to highlight the importance of correct hysteretic modeling in capturing the pounding response of multiple-frame bridges. The use of traditional models like the elasto-plastic and bilinear models can underestimate the severity of the pounding effect on the global frame response. CONCLUSIONS Recent earthquakes have illustrated that multiple-frame and multi-span simply supported bridges are vulnerable to damage from seismic pounding. Previous studies have typically used bilinear or stiffness-degrading models to represent the response of bridge columns while analyzing the effects of pounding. This paper investigates the inuence of column hysteretic characteristics such as stiffness degradation, strength deterioration, and pinching on the impact response of adjacent frames in a multiple-frame bridge. Traditional analytical models such as the elasto-plastic and bilinear models, and more advanced models such as the Q-Hyst stiffness-degrading and pivot strength-degrading models are considered for representing the response of bridge columns. A simplied planar nonlinear analytical model of a typical multiple-frame bridge is developed including inelastic frame action, nonlinear hinge behavior, and abutment effects. Parameter studies conducted on two adjacent bridge frames subject to ten far-eld earthquake records show that the traditional models underestimate the stiff frame displacement amplication due to pounding, and overestimate the exible frame amplication, when compared to the advanced models, for moderately to highly out-of-phase frames. At T1 / T2 = 0.3, the traditional models underestimate the stiff frame pounding response by 30% and overestimate the exible frame response by 20%. The effect of pounding is not signicant for in-phase frames T1 / T2 = 0.7 , irrespective of the hysteretic model considered. The strength degradation effect imposes no additional demands on the pounding response as long as stiffness degradation is modeled. However, in the presence of near-eld records, strength degradation increases the stiff frame displacement demand by 125% when compared to stiffness-degrading effect, for highly out-ofphase frames. Moreover, at T1 / T2 = 0.7, the displacement amplications due to pounding show greater discrepancy from unity for near-eld ground motions, with a stiff frame amplication of 1.3 and a exible frame deamplication of 0.85. A case study conducted on a four-frame bridge with the 1989 Saratoga record PGA= 0.5 g indicates that strength degradation in bridge columns combined with pounding can increase the stiff frame displacement response by 50%, when compared to other hysteretic models. The traditional models underestimate the stiff frame displacement, in good agreement with ndings from the earlier parameter study. In conclusion, the authors recommend the use of a stiffness-degrading model to accurately estimate the frame pounding response in the presence of far-eld ground motion records. The expected value of stiff frame displacement amplication for ground

1134

S. MUTHUKUMAR AND R. DESROCHES

motion effective period ratios in Zone I T2eff / Tg 1 is 1.7 with a standard deviation of 0.6 for highly out-of-phase frames T1 / T2 = 0.3 , and 1.4 standard deviation of 0.4 for moderately out-of-phase frames T1 / T2 = 0.5 . The above results are applicable for frame ductility demands around 4. In the case of near-eld ground motions, a strengthdegrading model is suggested for the bridge frame, to accurately estimate the pounding response. However, it is difcult to recommend a range for the frame displacement amplication factors, since the values are dependent on the strength degradation parameters selected and are highly sensitive to near-source records used in analysis. ACKNOWLEDGMENTS This study has been supported primarily by the Earthquake Engineering Research Centers Program of the National Science Foundation under Award Number EEC9701785 and the CAREER Program of the National Science Foundation under Grant 0093868. REFERENCES
Anagnostopoulos, S. A., 1988. Pounding of buildings in series during earthquakes, Earthquake Eng. Struct. Dyn. 16, 443456. California Department of Transportation Caltrans , 1999. Caltrans Seismic Design Criteria, version 1.1, Sacramento, CA. Clough, R. W., and Johnston, S. B., 1966. Effect of stiffness degradation on earthquake ductility requirements, in Proceedings, Japan Earthquake Engineering Symposium, Tokyo, Japan, pp. 195198. Davis, R. O., 1992. Pounding of buildings modeled by an impact oscillator, Earthquake Eng. Struct. Dyn. 21, 253274. DesRoches, R., and Fenves, G. L., 2001. Simplied restrainer design procedure for multipleframe bridges, Earthquake Spectra 17 4 , 551567. DesRoches, R., and Muthukumar, S., 2002. Effect of pounding and restrainers on seismic response of multiple-frame bridges, J. Struct. Eng. 128, 860869. Dowell, R. K., Seible, F., and Wilson, E. L., 1998. Pivot hysteresis model for reinforced concrete members, ACI Struct. J. 95, 607617. Earthquake Engineering Research Institute EERI , 1995a. Northridge Earthquake of January 17, 1994 Reconnaissance Report, Vol. 1, edited by J. F. Hall, Oakland, CA. Earthquake Engineering Research Institute EERI , 1995b. The Hyogo-Ken Nanbu Earthquake Reconnaissance Report, edited by C. D. Comartin, M. Greene, and S. K. Tubbesing, Oakland, CA. Earthquake Engineering Research Institute EERI , 2001a. Chi-Chi, Taiwan, Earthquake of September 21, 1999 Reconnaissance Report, edited by J. Uzarski, and C. Arnold, Oakland, CA. Earthquake Engineering Research Institute EERI , 2001b. The Nisqually, Washington, Earthquake of February 28, 2001Preliminary Reconnaissance Report, Oakland, CA. Earthquake Engineering Research Institute EERI , 2002. Bhuj, India, Earthquake of January 26, 2001 Reconnaissance Report, edited by S. K. Jain, W. R. Lettis, C. V R. Murty, and J. P. . Bardet, Oakland, CA.

EFFECT OF FRAME-RESTORING FORCE CHARACTERISTICS ON THE POUNDING RESPONSE

1135

Goldsmith, W., 1960. Impact: The Theory and Physical Behaviour of Colliding Solids, Edward Arnold, London, England, 379 pp. Jankowski, R., Wilde, K., and Fuzino, Y., 1998. Pounding of superstructure segments in isolated elevated bridge during earthquakes, Earthquake Eng. Struct. Dyn. 27, 487502. Kawashima, K., and Shoji, G., 2000. Effect of shock absorber to mitigate pounding effect between bridge decks, in Proceedings, International Workshop on Mitigation of Seismic Effects on Transportation Structures, National Center for Research on Earthquake Engineering, Taipei, Taiwan, R.O.C, pp. 207218. Kim, S. H., Lee, S. W., Won, J. H., and Mha, H. S., 2000. Dynamic behaviors of bridges under seismic excitations with pounding between adjacent girders, in Proceedings, 12th World Conference on Earthquake Engineering, Auckland, New Zealand, Paper 1815. Kim, S-H, and Shinozuka, M., 2003. Effects of seismically induced pounding at expansion joints of concrete bridges, J. Eng. Mech. 129, 12251234. Kreyzig, E., 1999. Advanced Engineering Mathematics, 8th edition, John Wiley & Sons, New York, pp. 942952. Kunnath, S. K., Reinhorn, A. M., and Park, Y. J., 1990. Analytical modeling of inelastic seismic response of R/C structures, J. Struct. Eng. 116, 9961017. Maison, B. F., and Kasai, K., 1992. Dynamics of pounding when two buildings collide, Earthquake Eng. Struct. Dyn. 21, 771786. Malhotra, P. K., 1998. Dynamics of seismic pounding at expansion joints of concrete bridges, J. Eng. Mech. 124, 794802. Muthukumar, S., and DesRoches, R., 2004. Evaluation of impact models for seismic pounding, in Proceedings, 13th World Conference on Earthquake Engineering, Vancouver, Canada, Paper 235. Naeim, F., and Kelly, J. M., 1999. Design of Seismic Isolated Structures: From Theory to Practice, John Wiley & Sons, New York, 289 pp. Pantelides, C. P., and Ma, X., 1998. Linear and nonlinear pounding of structural systems, Comput. Struct. 66, 7992. Saatcioglu, M., and Ozcebe, G., 1989. Response of reinforced concrete columns to simulated seismic loading, ACI Struct. J. 86, 312. Saiidi, M., and Sozen, M. A., 1979. Simple and Complex Models for Nonlinear Seismic Response of Reinforced Concrete Structures, Report No. UILU-ENG-79-2013, Structural Research Series No. 465, University of Illinois, Urbana. Shinozuka, M., Feng, M. Q., Kim, J-M., Nagashima, F., and Kim, H-K., 2000. Mitigation of seismic pounding effect on bridges using dynamic restrainers, in Proceedings of SPIEThe International Society for Optical Engineering, pp. 377387. Takeda, T., Sozen, M., and Nielsen, N., 1970. Reinforced concrete response to simulated earthquake, J. Struct. Div. ASCE 96, 5561. Yang, J. N., and Agrawal, A. K., 2002. Semi-active hybrid control systems for nonlinear buildings against near-eld earthquakes, Eng. Struct. 24, 271280.

Received 21 April 2004; accepted 31 January 2005