EXPERIMENTAL AND NUMERICAL ANALYSIS OF THE LATERAL RESPONSE OF FULL-SCALE BRIDGE PIERS

This paper presents the results of in-situ testing of two integrated pile-columns of a partially demolished bridge located in Auckland, New Zealand. A series of tests involving lateral monotonic pushover and subsequent dynamic free vibration snapback tests were used to quantify the variation in the stiffness and damping behaviour of the pile-column specimens over a range of lateral load levels. Each testing sequence consisted of incrementally increasing peak monotonic loads followed by the dynamic snapback, with a series of varying peak loads at the end of the testing sequence to evaluate the influence of loading history on the monotonic and dynamic response. The secant stiffness between the monotonic pushover tests performed to the same loading levels before and after the maximum load was applied, reduced by up to 40% in both the pile-columns, primarily due to soil gapping effects, highlighting the significant potential softening of the system prior to pile or column yielding. Progressive reduction in the damping of the system during each snapback test was evident, due to the varying contributions of different energy dissipation mechanisms, and the level of damping varied depending on the peak load applied. These results highlighted the significant influence of soil gapping and nonlinearity on the dynamic response of the system. Numerical models were developed in the open source structural analysis software OpenSeesPy using a Nonlinear Beam on Winkler Foundation approach to further investigate the response of the pile-columns. Models of both the pile-columns using existing p-y curves for clay soils showed good agreement with the experimental data in loaddisplacement, period and snapback acceleration time histories. Sensitivity analysis showed that the surface soft clay layer had a significant effect on the lateral response and dynamic characteristics of the model, reinforcing the need for good characterisation of the near surface soil profile to capture the behaviour of the system.


INTRODUCTION
The importance of accounting for soil-foundation-structure interaction (SFSI) in the seismic response of bridges has been widely acknowledged for many years [1][2][3][4][5]. An important part of the assessment of SFSI effects is the definition of the stiffness and damping characteristics of foundations. One of the inherent difficulties in accounting for these effects, particularly in a design setting, is this small number of large-scale pile foundations tests that have been performed relative to the wide range of structural and geotechnical characteristics that can be encountered in practice. This limitation means that there is often insufficient field test data to constrain the nonlinear stiffness and damping behaviour of the numerical models used to design and analyse pile foundations and wider bridge structures.
The lateral response of pile foundations has been previously studied using a range of different loading methodologies. The most widely used approach is static or pushover loading, where a gradually increasing load is applied to the pile head or at a location above the ground surface on an extension of the pile shaft. Using this approach, the load can be applied either monotonically or cyclically, with both methods typically applying loads of progressively larger magnitudes. Static loading of piles of various cross-sectional characteristics have been investigated in clay soils by Matlock [6] , Kramer [7], Rollins et al. [8], Gerber and Rollins [9], Sritharan et al. [10] and Sa'don [11], in sands by Chik et al. [12], Nasr [13], Madhusudan Reddy and Ayothiraman [14], Khari et al. [15], Abadie [16] and Aguirre et al. [17], in silts by Kramer [7] and Lalicata et al. [18], and in multi-layered profiles by Rollins et al. [19], Snyder [20] and Stewart et al. [21].
To capture the effects of damping in pile foundations, dynamic loading is required. In prior studies which have implemented dynamic loading, a range of excitations of increasing magnitude have been applied using an eccentric mass shaker, hydraulic actuator, or by quick release of the load at the end of a pushover loading cycle. Dynamic loading of piles of various crosssectional characteristics have been investigated in clay soils by Blaney and O'Neill [22], Crouse et al. [23], Boominathan [24], Sa'don et al. [25] and Fleming et al. [26], in sands by Scott et al. [27], Ting [28], Chai and Hutchinson [29], Boulanger et al. [30], Boominathan [24] and Chang and Hutchinson [31], in silts by Scott et al. [27], Ting [28], Boominathan [24], and in multilayered profiles by Boulanger et al. [32] and Boominathan [24].
Statnamic loading is another common testing method, where the duration of the load from the statnamic device is longer than a dynamic loading test and shorter than a static loading test, characterising a response that is somewhere between static and dynamic behaviour [33,34]. Statnamic loading of piles of various cross-sectional characteristics have been investigated in clayey soils by Brown [33,35], Brown et al. [36] and Brown and Hyde [37], in sands by Brown [33], El Naggar [34] and Tobita et al. [38], in silts by Brown [33], Brown et al. [39] and Tobita et al. [38], and in multi-layered profiles by Brown et al. [39], Bowles [40] and Broderick [41].
In general, the outcomes from previous research have provided designers with improved estimates of dynamic lateral stiffness and damping of pile foundations in several types of soil profiles along the depth of pile. Previous testing has enabled the development, evaluation and improvement of static and dynamic p-y curves for non-linear models and improved the existing theoretical equations for estimating SFSI effects. However, even with the significant number of large-scale pile tests in existing literature, there are few tests in soil profiles with soft surface layers underlain by much stiffer soils, as can often be found in the alluvial conditions at bridge river crossings. Due to the limited amount of in-situ test data, the design of piles in these soil conditions usually ignores the top soft layers if they are in locations prone to scour [42]. This paper presents the results of in-situ testing of integrated pile-columns of a partially demolished bridge located in Auckland, New Zealand. A test sequence involving static pushover and dynamic free vibration snapback tests was used to capture the change in dynamic response, stiffness and damping of the system due to pile cracking, soil gapping and soil nonlinearity. The characteristics of the bridge and the fieldtesting methodology are first presented, and the field testing results are discussed. To further explore the results from the field testing, a numerical model of each specimen was developed using the open-source structural analysis software OpenSeesPy [43]. Modelling results are compared to the field response and the sensitivity of key parameters controlling the response of the integrated pile-columns are evaluated and discussed.

FIELD TESTING SETUP AND METHODOLOGY
To investigate the performance of in-service bridge foundations, an experimental program was undertaken at Henderson Creek in Auckland, New Zealand. The Henderson Creek site consisted of two separate five span, reinforced concrete bridges that traverse a tidal creek, with this study focussing on Henderson Creek Bridge No. 2. This bridge was constructed in 1962, with an overall length of 93 m and a width of 9.35 m that supported two lanes of traffic. The superstructure of the bridge was supported by piers consisting of four 990 mm diameter reinforced concrete extended pile-columns with a centre to centre spacing of 2.24 m, connected by a 1.8 m x 1.09 m pier cap ( Figure 1). The above and below ground sections of the extended pile-columns had the same diameter, with the pile extending to a depth of 15.1 m, and no pile cap at the ground surface. The reinforcing steel layout is summarised in Figure 2 and consisted of twenty-four D32 longitudinal bars (32 mm diameter, fy = 300 MPa) and a spiral transverse reinforcement of 19 mm diameter (fy = 300 MPa) at 114 mm c/c. On the day of testing, the ground level was 6.55 m below the top of pile cap and water was present above the ground level during the testing.
The field testing programme focussed on the response of two of the extended pile-columns from one of the piers, referred to herein as Specimen 1 and Specimen 2. To isolate the test specimens the bridge deck and the longitudinal beams were removed, then the pile-columns were isolated by saw cutting through the pier cap at the locations shown in Figure 1, creating a clear gap of 1 m to prevent any interaction between the Specimens and other parts of the pier during testing.
The soil profile at the testing locations were characterised using a set of cone penetration tests (CPT) and rotary boreholes with Standard Penetration Tests (SPT). A total of four boreholes and eleven CPTs were used to characterise the soil profile at the test locations. These profiles are summarised in Figure 2. At the location of Specimen 1, a 4.5 m thick layer of soft alluvium was present at the ground surface, with a CPT tip resistance (qc) of 0.6 MPa and SPT N60 values of 2-8. Beneath this surface layer was a 1.5 m layer of highly-to-moderately-weathered East Coast Bays Formation (ECBF) sandstone and siltstone with a qc of 6 MPa throughout the layer. At the location of Specimen 2, an 8.1 m thick layer of soft alluvium was present at the ground surface, with qc of 0.4 MPa and SPT N60 values of 2-8. This surface layer was underlain by a 3 m layer of highly-tomoderately-weathered ECBF with qc of 5 MPa throughout the layer. The lower stratum for both Specimen 1 and Specimen 2 was slightly weathered ECBF with qc of 30 MPa, transitioning to unweathered ECBF over a depth of a few metres, with CPT refusal at this point. This unweathered ECBF formed the bearing layer for each specimen.

Testing Sequence
Testing was conducted to capture the change in dynamic response, stiffness and damping of the specimens. The testing of each specimen was performed in two parts: 1) a monotonic pushover (referred to as pushover herein) in which a horizontal load was applied slowly until a target load was reached and 2) a snapback in which the horizontal load was released at the target load, allowing the specimen to undergo free vibration. Table 1 summarises the loading sequence applied to each test specimen. A similar loading sequence was applied for both specimens, with a gradual increase in the target load up to Test 5, followed by a sequence of fluctuating target loads to assess the effect of past loading on the response.

Testing Methodology
The lateral load was applied to the top of each specimen using two pre-stressing strands that extended between steel loading frames attached to the top of the specimen and to the longitudinally adjacent bridge pier as shown in Figure 3 and Figure 4, similar to those used in bridge pier testing by Wood and Phillips [44]. The pier cap of the adjacent bridge pier was unaltered and acted as a reaction pier during testing. Each loading frame was hooked behind the back of the specimens and the reaction pier and secured with concrete anchors. Lateral load was applied 6.55 m above the ground level using a hydraulic actuator attached to the pre-stressing strands. The actuator was located inside the loading frame on the reaction pier and pushed a sliding block within the loading frames to tension the pre-stressing strands, as shown in Figure 4. The prestressing strands were installed with an initial slack to allow unrestrained movement of the specimens during free vibration. An electric pump was used to perform the static pushover test, increasing the horizontal load up to the target level for each pushover test. At the end of each pushover test, a snapback test was performed by dumping the hydraulic fluid from the actuator into a reservoir by remotely triggering a solenoid valve, creating a sudden release of load and allowing free vibration response.

Instrumentation
To measure the force-displacement response of the specimen during pushover testing, a Linear Variable Differential Transformer (LVDT) was positioned at the top of the specimen and connected to a reference frame supported by the adjacent pile. A load cell was connected to the actuator inside the loading frame at the reaction pier. To measure the dynamic response of the specimen during snapback testing, accelerometers were placed along the height of the specimen as indicated in Figure  3. An additional accelerometer was installed at the top of specimen in the direction perpendicular the loading direction to capture any out-of-plane accelerations. Throughout the duration of testing, the water level varied by up to a metre above the ground level, so it was not feasible to provide instrumentation near the ground surface. Test ID S1_T1 S1_T2 S1_T3 S1_T4 S1_T5 S1_T6 S1_T7 S1_T8 S1_T9

Static Pushover Testing
During the static pushover testing, specimens were pushed into the soil, and this resulted in plastic deformation on one side of the pile and opening of a gap on the other side as there was a separation between the pile and soil. At the end of a loading cycle when the load was released, there was a gap in front of the pile due to the plastic deformation of the soil, which along with the gap behind the pile reduced the stiffness of the system. This can be seen in the load-displacement response for Specimen 1 and Specimen 2 are shown in Figure 5. The loaddisplacement response for Specimen 1 is shown in Figure 5 (a) with the drift between the top of the pile and the ground level. Specimen 1 remained elastic during the first two low level pushover tests. A slight change in the secant stiffness was observed during S1_T3 at approximately 50 kN when the cracking moment of the pile section (Mcr = 310 kN-m) was exceeded in the critical sections of the specimen. At the start of S1_T5, there was a residual displacement of approximately 8 mm at the top of the specimen. This residual displacement can be attributed to: (1) small residual rotations at and below the ground level due to incomplete closure of cracks and the low axial load on the specimen [45]; (2) soil plastic deformation and the development of a gap between the pile and the surrounding soil during the previous pushover and snapback tests. While the gapping could not be observed directly due to the water level of the river, the load-displacement response in S1_T5 shows signs of gapping similar to other large scale pile test studies [1,26]. Namely, the secant stiffness at the beginning of the monotonic push was lower than the previous tests due to the loss of contact with the surrounding soil, and as the gap closed it put the pile in contact with a larger area of soil, increasing the stiffness of the system. After the gap fully closed, the S1_T5 curve aligned with the backbone curve from S1_T4. A similar behaviour of increasing secant stiffness at the later stages of the static pushover loading was observed in S1_T7. The loaddisplacement response of the specimen after S1_T5 remained linear, as all subsequent load levels following S1_T5 were less than the maximum load achieved in S1_T5, meaning it is unlikely additional gapping or soil nonlinearity developed.
The load-displacement response of Specimen 2 is shown in Figure 5 (b). The specimen response was elastic up to a load of approximately 30 kN during S2_T1. At the start of S2_T4 there was some residual displacement that developed through the same mechanisms described for Specimen 1 previously, and the load-displacement response was similar to that observed for Specimen 1 in S1_T5. A similar behaviour of increasing secant stiffness during gap closing was also observed in S2_T5. The load-displacement response of the specimen after S2_T5 remained linear as the load levels were less than the maximum load applied in S2_T5 and are similar to the response in the same tests in Specimen 1, except that 5 mm of additional residual deformation was observed following S2_T5. Note that Specimen 2 exhibited approximately half the stiffness of Specimen 1 during Tests 1-4 as well as larger residual displacement following Tests 4-6. These phenomena are likely due to more extensive cracking of the pile section prior to the start of testing and lower strength and stiffness of the soil layers at the location of Specimen 2 respectively.
Even though the load levels in Test 3 were twice that of Test 6 for both specimens, the peak displacements were very similar. This is a clear demonstration of the influence of gap formation between the pile and surrounding soil, reducing the overall stiffness of the specimen. For both the 100 kN tests (Test 3 and Test 8) there was a reduction in the secant stiffness of approximately 40%. The reduction in secant stiffness can be attributed to both an increase in soil gapping around the specimen and an increase in cracking of the pile section after the maximum target load was reached in Test 4 and Test 5. During the testing of both the specimens, the peak load levels were not large enough to exceed the yield bending moment of 1630 kN-m. The maximum load levels during the testing sequence were representative of the serviceability limit state design loads for this bridge.

Snapback Testing
Following the monotonic push to each target load in Table 1, a snapback test was performed to determine the dynamic properties of the specimens. The free vibration response of snapback tests for Specimen 1 and Specimen 2 are shown in Figure 6 and Figure 7 respectively. A summary of all the snapback test results is provided in an electronic supplement to allow for brevity here. For both specimens, the peak to peak amplitude decay during free vibration occurred at a relatively slow rate for tests with release loads lower than 100 kN (as illustrated by Figures 6a  and 6d, and 7a and 7d). In contrast, tests with release loads greater than 100 kN displayed a free vibration response with two distinct stages. Initially, there was a large reduction in amplitude during the first one to two cycles, which was followed by subsequent decay rates similar to that of the tests with lower initial release loads (as illustrated by Figures 6b and  6c, and 7b and 7c). The natural period of the two specimens was estimated for each snapback test using a combination of peak picking from Fast Fourier Transformation (FFT) and peak to peak intervals from the time series data. The resulting natural periods for each specimen are summarised in Figure 8. The range of natural period varied from 0.19 s to 0.24 s for Specimen 1 and 0.2 s to 0.29 s for Specimen 2. The natural period for Specimen 1 increased from 0.19 s to 0.22 s between tests S1_T1 and S1_T5 and then reduced to 0.2 s in S1_T9. Similarly, the natural period for Specimen 2 increased from 0.2 s to 0.29 s between test S2_T1 and S2_T5, and then reduced to 0.265 s in S2_T9. The natural period for both specimens increased as the peak load increased until the maximum peak load in Test 5, and then reduced with a reduction in peak load in subsequent tests. The natural periods were higher in tests after maximum peak load application compared to the same load levels before the application of the maximum peak load application in Test 5. This variation in natural period can be attributed to an increase in the formation of a gap between specimen and surrounding soil and specimen cracking. Tests with release loads above 100 kN had a 5% increase in the natural period between the initial rapid amplitude decay and the slow amplitude decay in later stages of the test. The range of natural periods for Specimen 2 were larger than the range of natural periods for Specimen 1 due to more significant cracking in the pile section in Specimen 2 prior to the start of testing and the lower strength and stiffness of soil layers surrounding Specimen 2.

Figure 8: Variation of the natural period for Specimen 1 and Specimen 2 with respect to the maximum load at release for each test.
The time series data in Figure 6 and Figure 7 indicate that damping was not consistent throughout the duration of the free vibration, being influenced by a combined mechanism including the impact between the pile and the surrounding soil and soil gapping around the pile. The contribution of each component to the total damping depends on the magnitude of the snapback release load and this contribution changes as the number of cycles of vibration increases.
To assess the variation in damping of each specimen throughout the response histories, equivalent viscous damping was determined using the logarithmic decrement method [46].
Multiple damping values were calculated from each test using sets of acceleration peaks from several stages of the response histories. This enabled a simplified equivalent viscous damping representation of the complex damping contributions from the previously discussed phenomena. The calculated equivalent viscous damping values are plotted in Figure 9 as a function of acceleration. The range of the maximum equivalent viscous damping estimates across all tests varied from 3.3% to 15% for Specimen 1 and 6% to 25% for Specimen 2. The maximum damping for Specimen 1 increased from 3.3% to 14% from S1_T1 to S1_T5 and then reduced to 6% in S1_T9. Similarly, the damping for Specimen 2 increased from 6% to 25% from S2_T1 to S2_T5, then reduced to 17% in S2_T9. This shows that the equivalent viscous damping values of 10% and 15% suggested by the New Zealand Transport Agency [47] for pier piles in granular and sandy soils and clay soils respectively provide a conservative estimate of the damping these components provide. Changes in the specimen properties meant that there was an increase in damping from the start to the end of the test sequence. The higher initial damping at low load levels in Specimen 2 compared to Specimen 1 could be due to the different soil profile characteristics and/or cracking of Specimen 2 prior to the start of testing. From Figure 9, it is clear there is a relationship between acceleration and the calculated equivalent viscous damping value. The maximum equivalent viscous damping in both specimens increased as the initial acceleration levels increased. The damping in subsequent tests were lower than the maximum viscous damping calculated for Test 5 but were higher than damping calculated for equivalent loads in tests conducted prior to Test 5. The high initial equivalent viscous damping for higher initial accelerations can be attributed to the dissipation of energy through the impact between the pile and the surrounding soil which occurs during the first half cycle of the free vibration response. The high initial equivalent viscous damping values would increase further when the specimen is pushed closer to the design level event demands. After this first half-cycle, the gap formation between the pile and soil influences the level of radiation damping of the foundation system. With the multiple mechanisms of energy dissipation that vary throughout the test and influence damping, it is not possible to represent damping throughout the entire response history using a single value of equivalent viscous damping. These variations in equivalent viscous damping were consistent with the observations made by other researchers [11,23].

FINITE ELEMENT MODELLING OF TEST BEHAVIOUR
The experimental results were used to validate numerical modelling procedures for integrated pile-columns using a Nonlinear Beam on Winkler Foundation approach. The opensource structural analysis program OpenSeesPy [43] was used to simulate the monotonic and dynamic snapback response of the integrated pile-columns tested in the experimental program whilst accounting for material and soil nonlinearity. This model calibration included the comparison of OpenSeesPy models with field tests in lateral load-displacements at the top of specimens during static pushover tests, the acceleration time histories at the top of specimens, and the variation in natural period of the specimen-soil system during free vibration. The validated numerical modelling procedures were then used to perform a sensitivity analysis study to identify the key parameters controlling the monotonic and dynamic response of the pile columns.

Model Overview
The integrated pile column models were modelled using displacement based distributed plasticity beam-column elements with three integration points. The column (above the ground level) was discretised into 40 equal length elements while the pile (below the ground level) was divided into 71 elements with smaller elements near the ground surface to capture the behaviour in the active pile length region. A convergence study was used to determine the number of elements within the column, as the natural period was sensitive to the number of elements above ground level. Mass corresponding to the tributary length of the pile-column cross section was lumped at the nodes between elements using a concrete density of 2400 kg/m3. An additional mass of 200 kg was lumped at the top of the column to account for test equipment. No P-delta effects were considered in the modelling as the peak drifts experienced during field testing were less than 1% and no additional axial load was applied to the test specimens.
Lateral loads were applied to the top of the column as static pushover loads. Force controlled loading was used to load all the specimens at the top of the specimen to replicate the field test loading conditions. All tests were modelled sequentially to ensure residual deformations in the system were captured. The snapback tests were modelled by releasing the lateral load and allowing free vibration. The rate of loading during all the tests was kept constant and the maximum load applied at the end of each cycle matched those summarised in Table 1.
Rayleigh damping was varied between different snapback tests, and within each snapback test based on the acceleration levels, to replicate the variation in damping observed in the field testing. Higher Rayleigh damping was used in the initial stages of each test to represent the damping when impact effects were dominant, and lower Rayleigh damping was used in the final stages of each test when radiation damping was dominant. During the testing, there was less than 0.1 m of free water above the ground level, such that any contribution to damping from the influence of the water surrounding the pile was likely to be negligible. The range of the initial damping estimates across all tests varied from 3.3% at initial accelerations of 2.8 m/s 2 , up to 25% for initial accelerations of 6.2 m/s 2 . Damping reduced down to 1% during the later stages of free vibration when the accelerations were below 0.5 m/s 2 . The target frequency used for defining the Rayleigh damping parameters were the natural frequency and three times the natural frequency for all tests. The Rayleigh damping provided to the system here was in addition to the hysteretic damping from soil and structural nonlinearity.

Pile-Column Modelling
The nonlinear behaviour of the pile-column was modelled using a fibre based approach with displacement based distributed plasticity beam-column elements. A convergence study was used to define the appropriate discretization of the fibre meshing, as it controlled the deflection of the pile-column post cracking. For every pile-column section, the core concrete fibre area was divided into 50 radial divisions and into 12 circumferential divisions, and the cover concrete area was divided into 20 radial division and 16 circumferential divisions. Based on the pushover test results, Specimen 2 had a lower initial stiffness which was attributed to initial cracking. To represent the effect of the reduced stiffness in the model, cracked concrete sections were considered in the pile active length region, which was up to 3m below the ground surface [50], as this region was most likely to have been cracked during the service life of the specimen. The reduced stiffness values were based on Paulay and Priestley`s recommendations for cracked concrete sections [2]. In this region, the initial cracked properties were applied to the entire cover region along with the exterior regions of the core concrete areas to achieve the recommended reduced stiffness values. The initial cracking was incorporated by using zero tensile strength in the concrete material model.

Soil Modelling
Soil was modelled using nonlinear p-y springs in OpenSeesPy [32] and were implemented using zero length elements with cyclic load-displacement relationships. The non-linear behaviour of the p-y spring is modelled using elastic, plastic and gap components in series. The gap component consists of a nonlinear closure spring in parallel with non-linear drag spring as shown in Figure 10. The p-y soil springs for soft alluvium, highly-to-moderately-weathered ECBF and slightly weathered ECBF soil layers were defined using recommendations from Reese and Welch [51] for over consolidated stiff clay. The ultimate capacity of the p-y springs (pult) were defined based on undrained shear strength derived from CPT tip resistance based correlations [52]. The displacement at which 50% of pult is mobilised (y50) were defined based on the pile diameter and the recommended range of strain50 values for Auckland clays (0.0015-0.002) from Sa'don et al [25]. The general formula for undrained shear strength can be written as: where su = undrained shear strength, qc = CPT tip resistance, σv0 = total overburden stress and Nk = empirical cone factor, that varies between 11 and 19.

Static Pushover Tests
The experimental and numerical load-displacement responses for each specimen are compared in Figure 11. In general, the numerical results compared well with the experimental observations for both specimens in terms of soil gapping, displacements at peak loads, and stiffness. Numerical and experimental results were similar up until S1_T3 for Specimen 1 in Figure 11 (a), with results diverging at loads greater than 100 kN. The lack of additional gap development during Tests 6-9 was also evident in the numerical results. The model for Specimen 2 in Figure 11(b) was able to capture the response up to test S2_T5. However, the numerical results for Specimen 2 did not accurately capture soil gapping following the maximum applied load during Test 5, as soil gapping was significantly underestimated by the model for Tests 6-9 (as shown in Figure  11). This may have been a result of the variability of the soil at the site, however further refinement of the model was not warranted.
To further explore the pushover response of each specimen, the variation of bending moment along the specimen height is presented in Figure 12. The location of the peak bending moment shifted downwards with an increase in the peak loading from Test 1 to Test 5 for both specimens, and varied between approximately 1-3 pile diameters below the ground surface for Specimen 1 and for Specimen 2 which is consistent with observations made by other researchers [17,29,44,53,54]. The location of peak bending moment in Specimen 1 in Figure 12 (a) shifted from a depth of 0.85 m for the 50 kN tests performed before peak loading (S1_T1 and S1_T2) to a 2.75 m depth for those tests performed after peak loading (S1_T6 and S1_T9). This is a result of gapping between the soil and pile, which influences the effective unsupported length of the integrated pile-column. A similar shift in the location of peak bending moment was observed between S1_T3 and S1_T8 and between S1_T4 and S1_T7. The peak bending moment locations for Specimen 2 in Figure 12 (b) showed a similar trend. These shifts in the bending moment profiles throughout the test sequence, and the differences between tests with similar load levels clearly demonstrates the influence of gapping around the pile and its importance in understanding foundation demands.

Snapback Tests
The snapback tests were used to evaluate the accuracy of the acceleration response predicted by the model. The experimental and numerical periods are shown in Figure 13 while the response histories from selected snapback tests have been plotted in Figure 14 (the full set of response histories is summarised in the electronic supplement). The natural period of the two specimens from OpenSeesPy models was estimated for each snapback test by identifying the peak in the Fast Fourier Transformation (FFT) of the time series data. In general, there is a good comparison between the experimental and numerical periods and dynamic response histories, with most of the periods from the numerical models within 10% of periods calculated from field testing. The models were able to replicate the snapback behaviour of both the specimens across all levels of loading. The model captured the acceleration response history well during the initial low level snapback tests of S1_T1 and during low level snapback test S1_T9, after the application of maximum loading in S1_T5. In both these tests the specimens have low initial acceleration and the acceleration decay was gradual throughout. The model also captured the acceleration response history well during the high level snapback tests of S1_T4 and S1_T5. In these tests the specimens had large initial accelerations with high acceleration decay due to the impact, followed by lower accelerations with gradual decay.

SENSITIVITY ANALYSIS
To evaluate the influence of key parameters on the pushover response of the integrated pile columns, a limited sensitivity analysis was conducted using Specimen 1 with the application of a 100 kN load as a reference. Specimen 1 was chosen for the sensitivity study as OpenSeesPy models were able to closely match the observed field testing response at this load level, including the lateral displacements, soil gapping, fundamental periods and acceleration time histories. For the loading range being assessed the parameters that were the focus of the sensitivity study were the initial tangent modulus of concrete, the ultimate capacity of the p-y springs (pult) and the displacement at which 50% of pult is mobilised (y50). The initial tangent modulus of concrete was modified by varying the concrete compressive strength, with values 76% to 125% of the reference value. This range is perhaps much larger than the overall uncertainty expected for structural properties, however the effects of ageing could account for the upper bounds of the uncertainty range, and the lower bound provides an analogue for accounting for the cracked stiffness of the pile-column. For each soil layer pult was varied directly, while the y50 value was modified by varying the strain50 value. The sensitivity to both these parameters was assessed using 50% to 200% of the reference value [55].

Figure 15: Effect of initial tangent modulus of concrete on (a) Specimen head displacements; (b) Peak bending moment profiles.
The influence of the initial tangent modulus of concrete on the pushover response is illustrated in Figure 15, with six different modulus values used to demonstrate the variation in response.
There was a 29% reduction in secant stiffness for a 24% reduction in the initial tangent modulus, while there was a 42% increase in secant stiffness for a 25% increase in the initial tangent modulus. As the initial tangent modulus increased, the rate at which the secant stiffness of the integrated pile-column increased slowed.
At some point further increases in the initial tangent modulus will have little effect on the specimen head deflections and system period, with the behaviour of the system dictated by the soil properties. The location of the peak bending moment was not influenced by the initial tangent stiffness of the concrete as indicated in Figure 15b.
The influence of the variation of soil properties for each soil layer on the pushover response is shown in Figure 16 and Figure  17 for pult and y50, respectively. In general, only the properties of the top soil layer (ground level to 4.5 m) influenced the static force-displacement response and bending moment distribution along the height, with minimal influence on response observed when the properties of the middle (4.5 m to 6 m from top of ground level) and bottom layers ( Similarly, there was a 10% increase in secant stiffness when the top layer y50 was reduced to 50% of the reference model while there was a 10% reduction in secant stiffness when the top layer y50 was increased to 200% of the reference model. The peak bending moment location shifted 0.1 m upwards towards the ground level from the reference peak moment location of 1 m when the top layer y50 was reduced to 50% of the reference model. The peak bending moment location moved 0.2 m downwards when the top layer y50 was increased to 200% of the reference model.

Figure 17: Effect of pult value of soil p-y springs on a) Specimen head displacements; b) Peak bending moment profile.
The combined results of the sensitivity analysis are summarised in Figure 18 with respect to the change in secant stiffness to highlight the controlling parameters. There clearly is significant sensitivity to changes in the concrete initial tangent modulus for the range considered. Only the pult and y50 parameters of the top soil layer influenced the response relative to the reference model, and the model remained insensitive to changes in pult and y50 values of the underlying soil layers. The importance of the soil strength and stiffness of the near surface soil layers on overall response are consistent with the observations made by other researchers [24,26]. The results in Figure 18 in terms of exact values are only directly applicable to Specimen 1, however the general trends related to which parameters control response would still be applicable to Specimen 2.

CONCLUSIONS
Static pushover and snapback tests were performed on two concrete integrated pile-column specimens of the Henderson Creek Bridge No. 2 to determine the static and dynamic behaviour of each specimen. The test sequence was chosen in order to capture the change in dynamic response, and variation in stiffness and damping of the specimen due to soil nonlinearity, gapping and cracking of the pile-column. The experimental results were used to validate a nonlinear model developed in OpenSeesPy, using a Nonlinear Beam on Winkler Foundation approach. Based on the field testing and modelling results, the following conclusions can be drawn:  Loading history has a significant influence on the secant stiffness of integrated pile-columns, with reduction in secant stiffness of up to 40% observed, primarily due to soil gapping around the piles.
 Damping during the dynamic response of integrated pilecolumns is influenced by a combination of mechanisms, including impact between the pile and surrounding soil and gapping. These mechanisms vary depending on the level of excitation and loading history, suggesting that constant equivalent viscous damping values should not be used when accounting for soil-foundation-structure interaction effects as part of detailed component modelling. For the wider analysis of bridges, the values suggested by NZ Transport Agency Bridge Manual for piles seems to provide a conservative estimate of the damping these components can provide.
 The existing p-y curves for clay type of soils can accurately predict the static and dynamic response of the integrated pile-columns, when appropriate soil and concrete material properties according to site conditions are provided.
 Even when the top soil layer consists of soft alluvial soils, the strength and stiffness of this layer has a significant effect on the lateral specimen head deflection, secant stiffness, period and the peak bending moment location of specimen-soil system. This reinforces the need for good characterisation of the near surface soil profile in order to capture the behaviour of the system.