Seismic soil pile interaction in liquefiable soil

Soil Dynamics and Earthquake Engineering 24 (2004) 551 564 www.elsevier.com/locate/soildyn Seismic soil pile interaction in liquefiable soil Assaf Klar, Rafael Baker, Sam Frydman* Faculty of Civil and Environmental Engineering, Technion, Israel Institute of Technology, Haifa 32000, Israel Accepted 21 October 2003 Abstract Numerical analysis of seismic soil pile interaction was considered in order to investigate the influence of flow mechanisms. Two models were employed a simplified model, where the pore pressure at any depth is that of the free field, and a more complete model in which the pore pressure is associated with three-dimensional flow. The soil behavior was modeled by a nonlinear, quasi-hysteretic constitutive relation. A parametric study was carried out, varying the superstructure mass and soil permeability. It was found that there is a pore pressure threshold below which both models yield similar results, but that this threshold cannot be quantified a priori, as it depends strongly on soil pile interaction. q 2004 Elsevier Ltd. All rights reserved. Keywords: Liquefaction; Piles; Lateral loads; Dynamic loads; Pile groups; Seismic response; Soil pile interaction; Nonlinear 1. Introduction Evaluation of seismic soil pile interaction in liquefiable soil is a complicated process involving a number of mechanisms that depend on the state and geometry of the system. Consequently, different approaches for analysis, experiments, and design related to this problem have been developed throughout the years. For example, in cases where the soil or phreatic surfaces are even slightly sloped, deformations of up to several meters may develop in cases where liquefaction occurs. In such cases, large lateral forces may act on the pile. This phenomenon is commonly referred to as lateral spreading and its effect on pile behavior is usually evaluated by a simplified method of applying static forces associated with the free field permanent displacement [1,2]. When lateral spreading does not occur, it may be possible to use nonlinear dynamic p y curves for analysis, but the nature of these curves will be dependent on the state of the soil. It has been observed in several shaking table tests, that once the soil liquefies locally, the subgrade reaction appears to be correlated with the relative soil pile velocity rather than with the relative displacement [3,4]. It has become quite common, in analyses of liquefiable soils, to use dynamic p ycurves which do not explicitly take * Corresponding author. E-mail address: cvrsfsf@tx.technion.ac.il (S. Frydman). account of pore pressure development [5,6]. This is legitimate if the permeability of the soil is either very small or very high; if it is high, no excess pore pressure is accumulated and the analysis is as for an equivalent dry sand; if it is low, the pore pressure which develops cannot dissipate, and its value is related, in some fashion, to the history of displacement, and therefore it can be implicitly taken into account by using degradation factors. Few successful attempts to correlate degradation factors with pore pressure ratio have been made [3,7]. However, these correlations should be used with caution, recognizing how each relation was derived, and for what purposes it can be used. For example, Dobry et al. [7] used shaking table centrifuge model tests to obtain a relation between pore pressure ratio and the degradation factor of p y curves. They applied seismic excitation for a limited duration while trying to maintaining a zero relative soil pile displacement condition. Immediately after the shaking, while there were still excess pore pressures in the soil, quasi-static cyclic lateral load tests were conducted. The relations between the p y curves and the pore pressure ratio were obtained by measuring instruments along the pile and miniature piezometers located far from the pile. From the description of the experiments, it is evident that use of Dobry et al. s correlation is legitimate for cases where excess pore pressure which develops around the pile due to soil pile interaction dissipates rather quickly, i.e. the representative pore 0267-7261/$ - see front matter q 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.soildyn.2003.10.006

552 A. Klar et al. / Soil Dynamics and Earthquake Engineering 24 (2004) 551 564 pressure is that of the free field. Whether this is always the case during real earthquake loading needs to be verified; part of the purpose of this paper is to consider this point. When conducting experiments with scaled models, similitude of flow conditions requires use of a fluid which is more viscous than water. Otherwise, the model is equivalent to a prototype with permeability N times higher than its true value (where N is the geometric scaling factor). Wilson [8] used centrifuge models for investigation of pile behavior under seismic loading in liquefiable sands using a viscous fluid. He observed that for a single pile, the excess pore pressure developed in the vicinity of the pile was quite similar to that developed under free field conditions. It should be noted that the fluid used was only viscous enough to reduce the scaled permeability from 30 to 3 times that of the prototype, and consequently significant dissipation of pore pressure in the vicinity of the pile may have occurred. Kagawa and Kraft [9] developed a numerical approach based on a nonlinear Winkler model, formulated using the concept of dynamic p y curves. They included a pore pressure generation model both due to free field response and the soil pile interaction, and compared their results with results from shaking table models. Unfortunately, they were unable to validate the soil pile interaction contribution to the pore pressure developed in their model, since the soil was liquefied by the free field pore pressure. Moreover, they did not incorporate redistribution of pore pressure generated by the soil pile interaction. Nevertheless, they suggested that the pore pressure build up, due to soil pile interaction, has a significant effect on the pile behavior. Kagawa [10] extended the model by including axisymmetric flow to represent the redistribution of the pore pressure generated by the soil pile interaction. Kagawa et al. [4] conducted shaking table tests on pile groups. They observed a distinct difference between excess pore pressures developed between the piles and those of the free field. They attributed this difference to reduction of dissipation of excess pore pressure, due to partial blockage of drainage by the piles. Finn et al. [11] used a finite element model for soil pile interaction, to simulate experimental result of piles in liquefiable soil. The model prevents motion in the vertical direction, and in the direction perpendicular to that of the pile motion. It appears that no excess pore pressure redistribution was considered in the study. The numerical simulation results showed excellent agreement with the experimental results, despite the restrictions on motion. Using numerical simulation, the present paper attempts to clarify the flow mechanism generated by soil pile interaction under seismic conditions, and to discuss its effect on the behavior of the pile. A parametric study was conducted for pile behavior using two assumptions: (1) the pore pressure at any depth is that of the free field, (2) the pore pressure results from 3D flow conditions. The analysis makes use of the following three partial models: (a) the numerical model which simulates the mechanics of the soil pile interaction, (b) a constitutive model describing the hysteretic behavior of the soil under cyclic loading, in terms of effective stresses, and (c) a model for generation and dissipation of excess pore water pressure. The following sections provide a more detailed description of each of these models. 2. Numerical model The numerical procedure, described by Klar and Frydman [12], for utilizing explicit, 2D, finite difference (FD) codes for approximate representation of 3D soil pile interaction, has been used in this study. They presented two models based on their procedure: a generalized Winkler (uncoupled) model, and a coupled model. In the coupled approach, used in this paper, the pile is represented by beam elements and the soil by an interacting, plane strain system. Klar and Frydman [12] verified their approach for the simple cases of dynamic loading of piles in homogenous, visco-elastic material. Fig. 1 shows the main elements of the procedure, in which the FD code, FLAC (Itasca [13]) was used. The technique is based on discretization of the 3D soil continuum into a series of horizontal layers, each represented by a 2D boundary value problem (BVP). A cavity is inserted in each layer to model the pile cross section. In addition, a separate grid, consisting of a series of connected, unsupported beam elements representing the pile, is defined. The calculation procedure advances in time, and develops the interaction between the two systems through transfer of velocities and forces from pile (beam) nodes to BVP grid points, and vice versa. In each calculation time step, the equations of motion of the pile are solved, and pile node velocities are applied to the corresponding cavity boundary nodes. Resulting forces acting on the cavity perimeter are calculated, and applied to appropriate pile nodes. These forces are then used to again solve the pile s equations of motion in the next time step. The coupled model, used in the present investigation, includes shear springs and dashpots joining every layer node to the corresponding nodes of the layer above and below. The derivation of interaction between the plane strain problems which was described by Klar and Frydman [12] was extended in the present investigation, in order to include a more general constitutive model. Consider the equations of motion for a continuous body: r 2 u j t 2 ¼ rg j þ s ij x i ð1þ where, u j is the displacement vector; r is the soil density; t is time; g j is the gravitation vector and s ij is the stress tensor. By restricting motion to the horizontal planes, only the variation of shear stress with depth needs to be considered for the coupling between adjacent plane strain problems. For homogenous, visco-elastic material, the coupling can be

A. Klar et al. / Soil Dynamics and Earthquake Engineering 24 (2004) 551 564 553 Fig. 1. Proposed procedure. (a) FLAC s original calculation cycle. (b) Modified calculation cycle. (c) Sub-grids and beam interaction. formulated directly by considering a body force, which is related to the displacement profile: f j ¼ s zj ¼ ð2g1! zj þ 2Gh _1 zj Þ ¼ G 2 u j z z z 2 þ h 2 _u j z 2 where, f j is the body force vector; G is the shear modulus; and h is the parameter related to the viscous damping. At the boundaries of the problem, different stress vectors are applied, according to the nature of the boundary (free, viscous, etc.). Since the plane strain problems overlie each other, Eq. (2) may be expressed, easily, by finite difference approximations. For a general constitutive model, a broader approach for coupling the plane strain problem should be invoked. The coupling process consists of the following three steps: (1) differentiation of the displacement profile at given time, t; in order to establish additional strain increments associated with the coupling between adjacent plane strain problems, (2) evaluation of new stresses corresponding to the new strains using the constitutive relation, (3) construction of body force vectors from these new stresses, based on Eq. (1). 3. Constitutive relation The constitutive relation considered is an extension of the quasi-hysteretic model described by Muravskii and Frydman [14]. The original model was formulated in terms of a system consisting of one degree of freedom bodies. In this model, stiffness and damping depend on the weighted mean (with respect to time) of strain and strain rate. The model of Muravskii and Frydman is here extended to 3D, ð2þ and expressed in terms of the following effective stress strain relations, which, for harmonic loading, result in elliptic hysteretic loops, with loss of energy independent of vibration frequency: " # _I 1 s 0 ij ¼ K I 1 þ h I I m 1 _ I m 1 I 1 ¼ 1 ii ; E ij ¼ 1 ij 2 I 1 3 d ij " # _E ij d ij þ 2G E ij þ h s E m ij _E m ij where I 1 is the first invariant of strains, E ij is the deviatoric strain tensor, K and G are the secant bulk and shear modulus, respectively, and h I ; h s are dissipative coefficients. Terms marked with superscript m in Eq. (3) are quantities averaged over time, according to the following expression, given by [14]: 2 3 ðt j n z 2 1=2 ðjþdj z m 0 ðtþ ¼6 4 t nþ1 7 ð4þ 5 n þ 1 where, j is an integration variable, and n is a parameter that allows varying relative weight to time (high n value assigns greater weight to recent value of z; i.e. the parameter n quantifies the notion of fading memory). Muravskii and Frydman showed that for a harmonic function, zðtþ; the average p ffiffi z m ðtþ approaches, with time, the constant value z 0 = 2 ; where z0 is the amplitude of the harmonic function, thus forcing Eq. (3) to result in elliptic hysteretic loops, with loss of energy independent of vibration frequency. It is generally accepted that energy losses in soils are independent of loading frequency, and the present constitutive model is consistent with this fundamental feature of soil behavior. ð3þ

554 A. Klar et al. / Soil Dynamics and Earthquake Engineering 24 (2004) 551 564 The parameters K; G; h I and h s in Eq. (3) should be functions of stresses and/or strains in order to account for soil nonlinearity. If they are set to be constant, the material will exhibit constant stiffness and material damping ratio. Various empirical relations have been suggested for stiffness and damping ratio under cyclic shear loading. Ishibashi and Zhang [15] collected experimental data from different sources and constructed a family of curves and expressions which incorporate the effects of confining pressure and shear strain on modulus reduction and damping ratio. Incorporation of Seed et al. s [16] expression for maximum stiffness in these expressions results in a representation of soil behavior under simple shear conditions. In the present investigation, these relations have been extended to a general state of stress and strain by considering quantities defined on the octahedral plane, resulting in the following expressions for dynamic shear modulus, G d ; and damping ratio, b s : G d ð s 0 ;g m s 0 0:5 octþ¼21:7k 2;max p a að s 0 ;g m octþ p a b s ð s 0 ;g m octþ¼0:333ð0:586a 2 ð s 0 ;g m octþ 2 1:547að s 0 ;g m octþþ1þ ð5þ where s 0 is the mean effective stress, p g m oct is average (with time) octahedral strain, g oct ¼ ffiffiffiffiffiffiffiffi 2=3J 2D ; J 2D ¼ 1 ij 1 ij = 2 2 1 ii 1 jj =6: K 2;max was used by Seed et al., and it depends on void ratio or relative density, p a is atmospheric pressure, and að s 0 ;g m octþ is the stiffness reduction curve of Ishibashi and Zhang, expressed in terms of g m oct : ( " að s 0 ;g m octþ¼0:5 1þtanh ln 0:000102! 0:492 #) m pffiffiffi 100 s0 12 g m oct p a ( " m ¼ 0:272 12tanh ln 0:000556! 0:4 #) pffiffiffi 12 g m oct ð6þ Muravskii and Frydman [14] showed that the functions G and h s in Eq. (3) can be obtained using the following expressions: qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi G ¼ G d ð s 0 ; g m octþ 1 2 ½2b s ð s 0 ; g m octþš 2 h s ¼ 2b s ð s 0 ; g m pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi octþ 1 2 ½2b s ð s 0 ; g m octþš 2 If soils were to exhibit uncoupled behavior for spherical loading and deviatoric loading, K and h I could be determined, directly, from cyclic spherical loading tests, and be related to s 0 and I 1 : Unfortunately, even assuming the uncoupling to be applicable, little experimental data can be found of cyclic spherical and deviatoric loading. If triaxial loading is considered, the ratio of radial strain to vertical strain may be referred to as minus the secant Poisson s ratio of the soil, n s ¼ 21 r =1 z : It can be shown ð7þ that in order for this ratio to be constant throughout a loading cycle, K must be proportional to G and h I must be equal to h s : Chen and Saleeb [17] discussed constitutive representation of nonlinear material using a nonlinear secant shear modulus and a secant Poisson s ratio. They pointed out that, due to experimental difficulties, it is hard to establish a reliable relation for the secant Poisson s ratio. Consequently, they suggested that it is satisfactory to assume a constant value of secant Poisson s ratio based on engineering judgment. Their suggestion is adopted herein; thus the ratio between K and G is 2ð1 þ n s Þ=3ð1 2 2n s Þ; while h I is equal to h s : It will be shown below that pile behavior is not sensitive to the choice of bulk modulus value or, alternatively, to the secant Poisson s ratio; thus engineering judgment is quite acceptable in this case. Furthermore, if no plastic volumetric strain results from the constitutive model, or if it is evaluated separately, under saturated undrained condition, the assumption of secant Poisson ratio has no significance as there is no volume change. 3.1. Application to seismic excitation If a dynamic system, such as a machine foundation, is excited from a state of rest to a steady state, a time-average value, as defined by Eq. (4), is appropriate. However, with increasing time the integral in Eq. (4) becomes large, and the sensitivity of the average value to strain changes decreases. Therefore, for earthquake excitation, which involves stochastic loading, Eq. (4) may be modified, as suggested by Muravskii and Frydman [12], by decomposing the overall excitation period into intervals of duration D; for each time interval, the average value is defined as follows: z m ðtþ ¼z m ðkdþð1 2 lþþ~z m l l ¼ðt2kDÞ=D 2 ðt 3 ðn þ 1Þ ðj 2 kdþ n z 2 1=2 ðjþdj ~z m kd ¼ 6 7 4 ðt 2 kdþ nþ1 5 where, D is the interval duration (time window length) and k the number of previous durations. For the first interval ðk ¼ 0Þ Eq. (4) should be used. This modification uses a moving time window, which allows a rapid change of average values. 4. Model for development and dissipation of pore pressure Under certain conditions, cyclic shear loading of sands may result in a tendency for volume reduction. For saturated sand the change of volume can occur only due to drainage. When drainage is not possible, the tendency for volume reduction will result in development of excess pore pressure, which will affect the behavior of the sand by ð8þ

A. Klar et al. / Soil Dynamics and Earthquake Engineering 24 (2004) 551 564 555 change of the effective stresses. On the other hand, in case of rapid drainage, effective stresses will not be affected and no significant generation of excess pore will occur. In fact, these two conditions are the limiting cases of a general situation where the drainage is controlled by the boundary conditions and the hydraulic conductivity of the soil. Numerical deformation analysis of undrained loading requires use of a value of total Poisson s ratio, n; close to 0.5, corresponding to the condition of zero volume change, or, alternately, an extremely high total bulk modulus, K; corresponding to incompressibility of water. The use of these high values leads to unacceptably small time steps, due to numerical stability requirements, and therefore should be avoided. In the case of analysis of the response of piles to dynamic loading, a justification for using lower values may be provided by the work of Jin et al. [18], who examined the lateral dynamic response of a pile in a linear, poroelastic material for a wide range of permeabilities. They concluded that the effects of permeability on stiffness and damping of the soil pile system are less than 10 and 35%, respectively. Since varying permeability would result in varying degree of drainage, and, as a result, to varying volume change in the soil, variation of permeability may be considered equivalent to varying n or K very low permeability would, for example, be represented by n close to 0.5 (or a high total K value), while a higher permeability would correspond to a lower n or K: Jin et al. s results suggest that specification of exact values of n and K is not essential in order to represent the mechanical response of the soil pile system. This conclusion is supported by numerical analyses conducted by Dobry et al. [19] which showed that if the horizontal pile stiffness is normalized using the soil elastic modulus, Poisson s ratio has little significance. In view of the above, it was decided that for the purpose of the present investigation, the mechanical response of the soil pile system would be analyzed, using a value of total n of 0.4 reasonably high, but low enough to prevent numerical problems, and to allow reasonable time steps. On the other hand, study of the pore pressure development and dissipation in the soil cannot be satisfactorily performed using a low value of n or K: Consequently, the calculation was decoupled, and a separate pore pressure development model was used for estimating pore pressure development and dissipation, as described below. 4.1. Pore pressure development Martin et al. [20] presented the following empirical relation which describes the incremental volumetric strain, due to a cycle of simple shear loading, on dry sand: D1 vd ¼ C 1 ðg 2 C 2 1 vd Þþ C 31 2 vd g þ C 4 1 vd where D1 vd is the increment of volumetric strain per cycle of loading, 1 vd is the accumulated volumetric strain and C 1 ; C 2 ; ð9þ Table 1 C 1 and C 2 constants [21] ðn 1 Þ 60 D r (%) C 1 C 2 5 34 1.00 0.40 10 47 0.50 0.80 20 67 0.20 2.00 30 82 0.12 3.33 40 95 0.06 6.66 C 3 ; C 4 are constants that depend on the relative density of the sand. Byrne and McIntyre [21] modified Eq. (9), and presented the following, simplified, expression for the incremental volumetric strain: D1 vd ¼ gc 1 exp 2C 21 vd ð10þ g Values of Byrne and McIntyre s constants, C 1 and C 2 ; are presented in Table 1 as a function of relative density and SPT blow count. The mechanism of pore pressure generation may (assuming applicability of superposition) be decomposed into two parts: (1) pore pressure development due to change in mean normal stress caused by the cyclic loading of the pile; (2) pore pressure development due to tendency for volume reduction ðd1 vd Þ caused by change in deviatoric or shear stress. Note that for a linear elastic material, this component of volumetric strain would be zero; consequently ðd1 vd Þ may be considered to be the plastic, volumetric strain resulting from deviatoric or shear stress. Pore pressure dissipation depends on permeability and flow boundary conditions. Since the seismic loading on the pile oscillates about zero, the first mechanism results in generation of excess pore pressure which also oscillates about zero, together with the mean normal stress, and, for undrained conditions (i.e. low permeability), does not change the mean effective stress. The second mechanism (i.e. pore pressure generated by deviatoric stresses) results in an excess pore pressure which accumulates with time, and does not oscillate around zero. If conditions (permeability, flow boundary conditions) are such that there is no significant pore pressure dissipation during an average loading cycle, the first mechanism has no effect on mean effective stress, which, consequently, changes with time only due to the second mechanism. That this is, in fact, the case, is suggested on the basis of the simplified, axisymmetric model for cyclic lateral loading of a pile considered by Martin et al. [22]. The model assumes incompressibility of water and soil particles, and refers to a steady state condition in which no pore pressures are developed due to tendency for compaction. Consequently, only the first mechanism is relevant. Using Martin s model, Fig. 2 has been developed, showing both the ratio of pore pressure, u; to the mean total normal stress, p; and the phase angle between u and p; as a function of nondimensional

556 A. Klar et al. / Soil Dynamics and Earthquake Engineering 24 (2004) 551 564 mean stress: D s 0 ¼ Dð s 0 2 s 0 0Þ¼2Du d ð11þ Fig. 2. Solution for steady state loading. period, T pc ; defined as T pc ¼ k=ðm v g w vr0þ; 2 where k is the soil permeability, m v is the soil compressibility, r 0 is the pile radius and v is the angular frequency of loading. Note that for zero change of effective stress, u=p should be unity, and the phase angle should be zero. It can be seen from Fig. 2 that as T pc decreases, the ratio u=p tends to 1.0 and that the phase angle diminishes (i.e. for small value of T pc the change in mean effective stress tends to zero). This behavior is particularly pronounced at large distances, r; from the pile. For earthquake loading and sand, the value of T pc is estimated to be in the range 0.01 1.0, for the range of confining stresses relevant to the effective depth of the pile. In this range of T pc ; Fig. 2 indicates that the effective stress would not be expected to change significantly due to the first mechanism. It seems reasonable to assume, then, that as long as the change of mean effective stress caused by the cyclic pile loading does not significantly alter the parameters k and m v ; the change in mean effective stress can be evaluated by considering only pore pressures developed by the second mechanism i.e. due to changes in deviatoric stresses. The above arguments are particularly relevant to the dynamic, soil pile interaction, in which there are increments of both spherical and deviatoric stresses. In addition, the effect of upward propagating seismic shear waves must be considered. However, these waves contribute only to the deviatoric cyclic part of pore pressure, and thus the above assumption is certainly justified with respect to this component. Based on the above, it is possible to obtain the following simple expression for the incremental change of the effective where s 0 0 is the initial (static) mean effective stress and Du d is the pore pressure increment generated by the deviatoric stress and modified by the dissipation process. Now, for undrained loading, the total increment of volumetric strain, D1 v ; must be zero. The total strain may be considered as the sum of the elastic, D1 e v; þplastic, D1 p v; components. As was pointed out earlier, deviatoric or shear stress results in a plastic volume strain increment D1 vd : Consequently this must be balanced by an elastic strain increment equal to 2ðD1 vd Þ: Substituting this value into the incremental linearelastic stress strain relations, changes in total and effective stresses may be calculated if an assumption is introduced regarding the relation between the normal strain increments. It is found that regardless of this assumption, the change in effective mean normal stress is equal to D s 0 ¼ 2K t D1 vd ; where K t is the tangent bulk modulus of the soil. This change in mean effective normal stress is used in Eq. (5) to adjust soil properties during the step by step analysis. On the other hand, the increments of total mean normal stress, and total excess pore pressure, are dependent on the assumed strain relationship. In the present investigation, it was assumed that all volume strain (D1 vd ; and volume strain resulting from pore pressure dissipation) results, solely, from vertical strain, i.e. D1 e x ¼ D1 p x ¼ D1 e y ¼ D1 p y ¼ 0: The assumption that volume strain is due mainly to vertical strain is supported by analyses of Davis and Poulos [23] who reviewed the differences and consequences caused by using the diffusion equation instead of the rigorous, Biot theory. They found that when the diffusion equation is solved under the assumption that the volumetric strain results only from vertical strains, the results are very similar to those obtained from Biot s theory. Although their discussion is related to loading of foundations, their conclusions support the present assumption. In addition, for undrained conditions, D1 e v ¼ 2D1 p v ¼ 2D1 vd : In the undrained case, the total pore pressure increment, Du; is found to be Du ¼ E r D1 vd ¼ðK þ 4=3GÞD1 vd ; and this value appears as the second term in Eq. (12), below, representing the pore pressure source in the diffusion equation. Use of the above model for pore pressure development has the advantage that it avoids the use of high values of Poisson s ratio, so significantly reducing the required time step for numerical stability. Using Darcy s law and considering both water and soil particles as incompressible, it is possible to derive the following differential equation relating the changes of pore pressure and volumetric strains: u t ¼ E r k x 2 u g w x 2 þ k y 2 u g w y 2 þ k z 2 u g w z 2! þ E r 1 vd t ð12þ

where E r is the tangent modulus of the one-dimensional unloading curve, equal to ðk þ 4G=3Þ; and k i is the permeability in the i direction. Eq. (12) (with the last term omitted) is the same as the Rendulic equation for multidimensional consolidation [24]. For simple shear conditions which exist far from the pile at the free field, Eq. (12) degenerates into the relation suggested by Finn et al. [25]. Martin et al. [20] suggested an analytical expression for the rebound modulus, which can be written in the following dimensionless form: A. Klar et al. / Soil Dynamics and Earthquake Engineering 24 (2004) 551 564 557 E r ¼ P a ðs 0 v=p a Þ 12m mk 2 ðs 0 v0 =P aþ n2m ð13þ where, parameters m; k 2 ; and n may be obtained from three unloading curves from different initial vertical stresses s 0 v0: On the basis of data presented by Martin et al. [20], and Seed et al [26], values of the parameters in Eq. (13) were taken as follows: for D r ¼ 45%; n ¼ 0:62; m ¼ 0:435; and k 2 ¼ 0:0028; for D r ¼ 60%; n ¼ 0:62; m ¼ 0:45 and k 2 ¼ 0:00167: Both the relations of Martin et al. (Eq. (9)), and Byrne and McIntyre (Eq. (10)) relate the volumetric strain to a shear strain associated with one directional shaking. For the present problem, these relations should be extended to the general case of multidirectional shaking. Considering a deviatoric strain space, G ¼ 1 1 2 1 p ffiffi 2 ; 1 1 2 1 p ffiffi 3 ; 1 2 2 1 T p ffiffi 3 ; 1 12 ; 1 13 ; 1 23 ; 6 6 6 p the octahedral shear strain is equal to g oct ¼ ffiffiffiffi 2=3 lgl: Using g oct as an indicator for cyclic shear strain amplitude will be efficient only in the case where all components of G are oscillating around zero with the same frequency and phase angle. In the general case, which includes, both, offset in shear strain and a combination of frequencies, it is necessary to monitor the magnitude, lgl; of G: The first reference point, G R ; is taken as the origin of the octahedral plane. Each time lg 2 G R l reaches a maximum value the reference point is reset to the value of G and it is regarded as an unloading point. The engineering shear strain amplitude is calculated as the distance from the new unloading point to the previous one. An identical strain space concept was presented by Itasca [13], but their search for maximum distance utilizes the projection of G 2 G R on the previous direction of G 2 G R : In the present analyses, half of the D1 vd value obtained by Eq. (10) is added gradually, over an interval of time equal to half the time from the previous unloading point to the new one. This is with the intention to simulate the significant volume change which occurs during unloading portions of the cycle, as observed by Finn [25]. The above pore pressure generation model and effective stress constitutive model were used in numerical analyses of laboratory liquefaction tests performed by De Alba et al. [27] and Finn et al. [25]. A comparison of the experimental Fig. 3. Numerical liquefaction resistance analysis. and numerical results (Fig. 3) shows good agreement between the two. 5. Numerical analysis 5.1. Purpose and input A parametric study was conducted for the case of a pile embedded in a homogeneous sand layer overlaying a rigid stratum. The water table is located at a depth of 1.5 m. A 0.5 m 2 cross section pile, with Young s Modulus of 3 10 7 kpa pinned at its tip was considered. A preliminary study was conducted for the site response alone, in order to choose a suitable earthquake excitation. The purpose of the present study was to investigate the conditions under which the pore pressure development is dominated by soil pile interaction, rather than by site response affects. In order to achieve this, it was decided to consider an earthquake satisfying the following requirements: (1) the dominant frequency of the excitation is in the valley between the first two maxima of the site amplification spectra, and (2) the dominant frequency of the excitation is not far from the resonance frequency of the pile superstructure system. For linear elastic soil, with stiffness increasing proportionally to square root of depth, the fundamental frequency of the soil deposit may be estimated by f s ¼ 0:223V SH =H [28], where V SH is the S wave velocity at depth H (bottom of the soil layer). Employing this expression, with the small strain shear wave velocity, results in upper bounds on natural frequencies of the site. For sands with relative densities in the range D r ¼ 45 60% the first two natural frequencies are approximately 3.5 and 9.5 Hz. The nonlinear behavior of the soil may decrease these natural frequencies somewhat, but the preliminary parametric study showed that this shift is not very significant, provided that liquefaction does not occur. The E W record of the October 1, 1987 Whittier earthquake, whose time history is shown in Fig. 4, had a dominant frequency in the range 5.5 7 Hz (Fig. 5), thus satisfying requirement (1) above, and

558 A. Klar et al. / Soil Dynamics and Earthquake Engineering 24 (2004) 551 564 Fig. 4. Acceleration bedrock input data. this record was chosen as input for the present analyses. The second requirement can always be satisfied by an appropriate choice of the superstructure mass; thus it is not a limiting factor with respect to the choice of the earthquake. The resonant frequency, f ; of the pile soil superstructure system is, approximately, inversely proportional to the square root of the superstructure mass M; i.e. it is possible to write: f ðhzþ ¼a=½MðtonÞŠ 0:5 ; where a is a constant of proportionality. A preliminary parametric study, considering sand with relative density of D r ¼ 60% showed that under small amplitude impact loading this constant is equal to approximately 70. Decreasing the soil pile stiffness by 50%, a is reduced to approximately 50. For the range f ¼ 5:5 7 Hz, and a ¼ 50 70; with the appropriate range of M is found to be 50 160 ton. For sandy soils, the coefficient of permeability has been taken in the range 10 21 10 23 cm/s. 5.2. Free field analyses Figs. 6 and 7 show the development of excess pore pressure with time in the free field (i.e. distant from the pile), for sand layers of different relative densities, permeabilities and bottom drainage conditions. For a given loading, the liquefaction potential depends on both permeability and relative density. It can be seen that the lower the permeability the less influence the drainage conditions have on the results. Note that the model does not take account of change in permeability which may result from change in effective stress. Since liquefaction is associated with a complete loss of stiffness, so that E r approaches zero, Eq. (12) shows that, at liquefaction, if the permeability, k; is constant, the rate of pore water pressure dissipation, u= t; approaches zero. To avoid such situations, pore pressure was prevented from increasing above 99% of the initial, effective vertical stress. For the range of permeabilities relevant for sands, and the geometric configuration examined, it was found that the drainage condition at the base of the layer has little effect on the response of the upper part of the profile, which is of major interest for soil pile interaction. This is, apparently, a result of the fact that the effective stress changes at the bottom of the layer are too small to significantly affect mechanical behavior, while the permeability is sufficiently small so that pore pressures in the upper zone are relatively Fig. 5. Fourier spectrum of acceleration input data.

A. Klar et al. / Soil Dynamics and Earthquake Engineering 24 (2004) 551 564 559 Fig. 6. Excess pore pressure development (no bottom drainage). independent of bottom drainage conditions. Consequently, only cases of impermeable base were further considered in the present investigation. 5.3. Analyses of soil pile interaction Although use of the pore pressure model in high permeability soil (k ¼ 10 21 cm/s) is legitimate for site response analysis, where no change in total normal stresses occurs, it may not be adequate for soil pile interaction analysis, since T pc (Fig. 2) may be too high. Consequently, no soil pile interaction analyses were carried out for this permeability. Furthermore, since a quick liquefaction occurs in the free-field sand layer with relative density of 45% for lower permeabilities, soil pile interaction analysis was not performed for that relative density. The effect of pore pressure development and dissipation on soil pile interaction was studied by comparing the results of two different models. The first model (model 1D) is based on the assumption that pore pressure at any depth is defined by free field conditions, involving onedimensional, vertical flow. The second model (model 3D) takes into account the pore pressure generation and distribution according to the 3D flow equation (12), which includes the influence of the soil pile interaction. For discussion purposes, it was found informative to separately consider two portions of the chosen earthquake those preceding and following 6 s; the first portion

560 A. Klar et al. / Soil Dynamics and Earthquake Engineering 24 (2004) 551 564 Fig. 7. Excess pore pressure development (bottom drainage). includes the strongest vibrations. It is convenient to represent the response of the pile in terms of the maximum bending moment developed in it; Figs. 8 and 9 show the maximum moment, M max (normalized by superstructure weight, W; and pile width, B) in the pile in these two time periods, as a function of depth, for permeabilities 10 22 and 10 23 cm/s, respectively. The following observations appear to be significant: (a) In the case of the 1D model, maximum moment distributions appear similar for both permeabilities. This suggests that the difference between the free field pore pressure distributions resulting from the different permeabilities (Fig. 6) is too small to (b) significantly influence the loading of the pile and the soil pile superstructure interaction. On the other hand, in the case of the 3D model, the moment distributions are not independent of permeability, indicating that soil pile interaction and 3D flow conditions lead to significantly different pore pressure distributions for the different permeabilities. Comparison of the moments resulting from the two models indicates that: (i) For the higher permeability (Fig. 8), both models indicate similar moment distributions during the second (weaker) portion of the excitation, for all considered superstructure masses, and during the first (stronger) excitation period for the smaller

A. Klar et al. / Soil Dynamics and Earthquake Engineering 24 (2004) 551 564 561 Fig. 8. Maximum moment distribution (k ¼ 10 22 cm/s). (ii) (75, 100 ton) superstructure masses. For the larger masses (125, 150 ton), the distributions during the first (stronger) excitation period were different for the two models, but similar to those obtained for the lower permeability. For the lower permeability (Fig. 9), the maximum moment distributions based on the two models are different, for all superstructure masses. These observations appear to indicate that differences between the models resulted, effectively, form cases where large pore pressures developed (due to stronger earthquake excitation, stronger soil pile interaction as a result of larger superstructure inertial effects, lower soil permeability), while the models predicted similar soil pile interactions in cases where lower pore pressures develop. The significance of the above observations may be better understood by reference to the evolution of moment with time, shown in Figs. 10 12. Figs. 10 and 11 refer to the case of the higher permeability, k ¼ 10 22 cm/s; it can be seen that for the 100 ton superstructure, the two models result in similar values and trends. On the other hand, for the 125 ton superstructure, significant deviation in results occurs at approximately 4 s (corresponding to the peak excitation pulse see Fig. 4). This deviation dies out at the end of the first excitation period. It is observed in Fig. 4 that the last part (about 5 6 s) of the first excitation period consists of decreased accelerations. It would appear that the duration of this decreased acceleration portion is sufficiently long so as to allow dissipation of excess pore pressure, thus resulting in similar conditions for both models. Furthermore, the second excitation period is either too short or too weak to generate, and accumulate, excess pore pressures sufficiently large to cause the predictions of the models to deviate from each other. Different observations are obtained for the low permeability case (Fig. 12). In contrast to the previous case, a distinct difference in values and trends is observed already for 100 ton superstructure. In addition, it seems that for low permeability, the duration of the low acceleration part of the first excitation period is not sufficiently long for the excess pore pressure to dissipate. From the fact that there were some cases in which the two models generated similar results, and other cases in

562 A. Klar et al. / Soil Dynamics and Earthquake Engineering 24 (2004) 551 564 Fig. 9. Maximum moment distribution (k ¼ 10 23 cm/s). which similar results were obtained by the 3D model for different permeabilities, it is suggested that there is a pore pressure threshold which causes a change in mechanical behavior. This is supported by inspection of the behavior of a single soil element under uniform, stress controlled cyclic loading, as predicted by the present constitutive relation. Fig. 13 shows the pore pressure ratio (the ratio of excess pore pressure to initial vertical effective stress), and Fig. 10. Moment with time. Mass, 100 ton; k ¼ 10 22 cm/s. Fig. 11. Moment with time. Mass, 125 ton; k ¼ 10 22 cm/s.

A. Klar et al. / Soil Dynamics and Earthquake Engineering 24 (2004) 551 564 563 Fig. 12. Moment with time. Mass, 100 ton; k ¼ 10 23 cm/s. the cyclic shear strain ratio (the ratio of shear strain to that of the first cycle) as a function of number of cycles, N; normalized by N L (the number of cycles required for liquefaction). It can be seen that the development of cyclic shear strain ratio can be roughly separated into two modes (illustrated by the dashed lines). The first mode is related to a pore pressure ratio smaller than about 75%, and is expressed in a decrease of stiffness of up to 50%. The second mode is related to pore pressures higher than 75%, and it results with large to complete loss of stiffness. This threshold pore pressure ratio, which may be considered a stability threshold, is related to the specific numerical simulation conducted, and it should not be taken as a representative value for all cases. Nevertheless, similar behavior can be observed, in results of many reported experimental investigations, when the stress path crosses the phase transformation line [29]. It is suggested that similar behavior exist for the soil pile interaction problem, i.e. as long as the accumulated pore pressure does not pass the pore pressure threshold, the change of stiffness is not large enough to create a distinct difference in response predicted by the simpler 1D, and the 3D pore pressure models. However, once the pore pressure passes this threshold, the behavior mode is changed and the two models generate different responses. Whether the pore pressure reaches the threshold value depends on the rate of pore pressure accumulation and dissipation, and so, on the soil permeability and superstructure mass. Consider, for example, the case of a superstructure mass of 150 ton. As a result of the high acceleration excitation during the first portion of the excitation, the pore pressure passes the threshold value both in the low and the high permeability analyses. Consequently, similar values of maximum moment are observed in the 3D flow analyses. However, for the second, weaker, excitation period, the pore pressures, in the high permeability case, dissipate to values below the threshold value, and approach the free field values. Consequently, in this case, the behavior is similar to that predicated by the simpler, 1D model in which the pore pressures are assumed to be the free field values. This threshold approach can be supported by experimental and analytical work reported by Fukutake et al. [30] for soil structure interaction. They showed that analysis using a total stress approach leads to good agreement with both experimental results, and analytical results based on an effective stress approach, for excess pore pressures ratios smaller than 70%. These observations suggest that as long as the effective stress state is far from the phase transformation line, the exact pore pressure distribution is not significant, and it is possible, therefore, to base the calculation on the free field pore pressure distribution, rather than on a more realistic one. 6. Summary and conclusions Fig. 13. Mechanical behavior in liquefaction resistance test. The seismic response of a single pile in a liquefiable sand layer was numerically evaluated using two models a simplified model, where the pore pressure at any depth is that of the free field, resulting from 1D flow, and a more complete model in which the pore pressure is associated with 3D flow. In some cases, both models generated similar results, while in other cases significant differences were observed. It is suggested that the similar response predicted by the two models corresponds to cases in which the pore pressure developed near the piles is below a threshold value. If pore pressures exceed this threshold value, the more complete, 3D model should be employed. It is suggested that this pore pressure threshold corresponds to the point at which the effective stress path meets the phase transformation line; its value cannot be quantified a priori;

564 A. Klar et al. / Soil Dynamics and Earthquake Engineering 24 (2004) 551 564 however, it is related directly to the soil pile interaction. The development of excess pore pressure is highly dependent on the soil pile superstructure interaction, its magnitude and frequency, while the dissipation is dependent on the permeability, drainage conditions, and excess pore pressure distribution (which is also related to the excess pore pressure development). Undoubtedly, the higher the permeability, the greater the likelihood that the pore pressure will not pass the threshold value, and the models will yield similar results. It is also suggested that a simplified pile soil model such as a Winkler model which use only the pore pressure of the free field for analysis, may be practical in highly permeable soils. However, straightforward guidelines to define when such simplified approaches are suitable could not be formulated from the present numerical analyses. Acknowledgements The research described in this paper is being supported by the Israel Ministry of Housing and Construction, through the National Building Research Station at the Technion. References [1] Stewart HE, O Rourke TD. The effects of liquefaction-induced lateral spreading on pile foundations. Soil Dyn Earthquake Engng 1991; 10(5):271 9. [2] Wang ST, Reese LC. Design of pile foundation in liquefied soils. Geotech Spec Publ 75, ASCE, Geotech Earthquake Engng Soil Dyn III 1998;1331 43. [3] Tokimatsu K, Zusuki H, Suzuki Y. Back-calculated p y relation of liquified soils from large shaking table tests. Fourth Int Conf Recent Adv Geotech Earthquake Engng Soil Dyn 2001; Paper 6.24. [4] Kagawa T, Minowa C, Abe A, Oda S. Shaking-table tests on and analyses of piles in liquefying sands. Earthquake Geotechnical Engineering Proceedings of the First International Conference, IS- Tokyo 95; 1995. p. 699 704. [5] Boulanger RW, Curras JC, Kutter BL, Wilson DW, Abghari A. Seismic soil pile structure interaction experiments and analyses. J Geotech Geoenviron Engng, ASCE 1999;125(9):750 9. [6] Curras CJ, Boulanger RW, Kutter BL, Wilson DW. Dynamic experiments and analyses of a pile-group-supported structure. J Geotech Geoenviron Engng, ASCE 2001;127(7):585 96. [7] Dobry R, Taboada T, Liu L. Centrifuge modeling of liquefaction effects during earthquakes. In: Ishihara K, editor. Earthquake geotechnical engineering. Rotterdam: Balkema; 1997. p. 1291 324. [8] Wilson DW. Soil pile superstructure interaction in liquefying sands and soft clay. PhD Dissertation, Department of Civil and Environmental Engineering, University of California at Davis; 1998. [9] Kagawa T, Kraft LM. Lateral pile response during earthquakes. J Geotech Engng, ASCE 1981;107(GT12):1713 31. [10] Kagawa T. Effect of liquefaction on lateral pile response. ASCE Geotech Spec Publ 34, Piles under Dynamic Loads 1992;207 23. [11] Finn WDL, Thavaraj T, Wison DW, Boulanger RW, Kutter BL. Seismic analysis of piles and piles groups in liquefiable sand. Seventh Int Conf Numer Models Geomech, Numog 1999;287 92. [12] Klar A, Frydman S. Three-dimensional analysis of lateral pile response using two-dimensional explicit numerical scheme. J Geotech Geoenviron Engng, ASCE 2002;128(GT9):775 84. [13] Itasca. FLAC user manual, Itasca Group, Minneapolis; 2000. [14] Muravskii G, Frydman S. Site response analysis using a non-linear hysteretic model. Soil Dyn Earthquake Engng 1998;17:227 38. [15] Ishibashi I, Zhang X. Unified dynamic shear moduli and damping ratios of sand and clay. Soils Found 1993;33(1):182 91. [16] Seed HB, Wong RT, Idriss IM, Tokimatsu K. Moduli, and damping factors for dynamic analysis of cohesionless soils. J Geotech Engng, ASCE 1986;112(GT11):1016 32. [17] Chen WF, Saleeb AF. Constitutive equations for engineering materials. Elasticity and modeling, New York: Wiley; 1982. [18] Jin B, Zhou D, Zhong Z. Lateral dynamic compliance of pile embedded in poroelastic half space. J Soil Dyn Earthquake Engng 2001;21:519 25. [19] Dobry R, Vicente E, O Rourke MJ, Roesset M. Horizontal stiffness and damping of single piles. J Geotech Engng, ASCE 1982;108(GT3): 439 59. [20] Martin GR, Finn LWD, Seed HB. Fundamentals of liquefaction under cyclic loading. J Geotech Engng, ASCE 1975;101(GT5):423 38. [21] Byrne PM, McIntyre J. Deformation in granular soils due to cyclic loading. ASCE GSP 40. Vertical and horizontal deformations of foundation and embankments; 1994. p. 1864 96. [22] Martin GR, Lam IL, Tsai CF. Pore-pressure dissipation during offshore cyclic loading. J Geotech Engng Div 1980;106(GT9): 981 96. [23] Davis EH, Poulos HG. Rate of settlement under two- and threedimensional conditions. Geotechinque 1972;1:95 114. [24] Lee IK, White W, Ingles OG. Geotechnical engineering. Pitman Publishing Inc.; 1983. [25] Finn WDL, Lee KW, Martin GR. An effective stress model for liquefaction. J Geotech Engng, ASCE 1977;103(GT6):517 33. [26] Seed B, Pyke RM, Martin GR. Effect of multidirectional shaking on pore pressure development in sands. J Geotech Engng, ASCE 1978; 104(GT1):27 43. [27] De Alba P, Seed HB, Chan CK. Sand liquefaction in large-scale simple shear tests. J Geotech Engng, ASCE 1976;102(GT9):909 27. [28] Gazetas G. Foundation vibrations. In: Fang H-Y, editor. Foundation engineering handbook, 2nd ed. New York: Van Nostrad Reinhold; 1991. p. 553 93. [29] Ishihara K. Stability of natural deposits during earthquakes. Proc 11th Int Conf Soil Mech Found Engng, San Francisco 1985;1:321 76. [30] Fukutake K, Ohtsuki A, Sato M, Shamoto Y. Analysis of saturated dense sand-structure system and comparison with results from shaking table test. Earthquake Engng Struct Dyn 1990;19:977 92.