Beam on Winkler foundation method for piles in laterally spreading ...

BEAM ON WINKLER FOUNDATION METHOD FOR PILESIN LATERALLY SPREADING SOILSAkihiro Takahashi, M.ASCE 1 , Hideki Sugita 1 and Shunsuke Tanimoto 1ABSTRACTThe paper presents application of beam on non-linear Winkler foundation methodto piles in laterally spreading liquefied soils. In many cases, parameters used in themethod are determined to fit particular case studies and are hardly applicable to theother separate cases, due to difficulties in the modelling of complicated phenomenawith the simplified method. In this study, variation of p-y curve parameters for pilesin laterally spreading liquefied soils is systematically examined by numerically analysingphysical model tests undertaken by independent researchers.INTRODUCTIONOne of the major sources of earthquake-induced damage to pile foundations is lateralspreading of liquefied soils. In practice, to assess performance of piles subjectedto kinematic loading due to large ground deformation, the beam on non-linearWinkler foundation method is used in simplified design procedures and has been introducedin some of practical design codes in Japan (JSCE 2000). This involvespseudo-static analysis of a pile foundation subjected to a superstructure inertial force(if applicable) and soil movement through Winkler springs that model soil-pile interaction.When this technique is applied to a problem on a foundation in laterallyspreading soils, accumulation of displacements in cyclic loading may be modelled in‘monotonic’ way, which involves considerable uncertainties and makes it difficult todetermine appropriate parameters for non-linear Winkler springs (p-y curves).1Public Works Research Institute, Tsukuba, Japan.1

In many cases, determination of parameters used in the method are made using thelimited number of physical model test simulations or case studies and are hardly applicableto the other separate cases (e.g. Brandenberg et al., 2001), due to difficultiesin the modelling of complicated phenomena with the simplified method. For instance,effects of cyclic loading and nature of input earthquake motion cannot be explicitlyconsidered in the monotonic loading numerical analysis. This kind of effectsin the monotonic loading analysis has not been fully understood yet and is hidden inthe model parameters.To assess sensitivity of parameters for the method on foundation responses in laterallyspreading liquefied soils, many case studies are required. For such purpose, bothfield observations and physical model tests may be utilised with careful regard totheir advantages and limitations: Data themselves are very useful but always involveuncertainty in cause of events for the former, while foundation responses during anearthquake are clear but the tests are always conducted under unpreferable sideboundary conditions for the latter. When prediction methods for ground deformationare assessed, the former may be preferable, while it is the latter for the non-linear p-ycurve parameters. In this study, variation of p-y curve parameters for piles in laterallyspreading liquefied soils is systematically examined by numerically analysing physicalmodel tests undertaken by independent researchers using the beam on non-linearWinkler foundation method with hyperbolic type p-y curves.PHYSICAL MODEL TEST RESULTS USEDThe target situation is that a pile foundation located in loose sand deposit behindgravity type quay wall moves seaward during earthquake due to lateral liquefied soilspreading initiated by quay instability. Schematic view of the problem to be solved isillustrated in Fig. 1. Data of physical model tests on the target problem were collectedfrom well-documented technical papers and reports.Summary of the physical model tests used are listed in Table 1 in ascending orderof the model pile diameter. Data of six different series of physical model tests areused: Half of them are ordinary shaking table tests and the others are dynamic centrifugetests. (All the values for the centrifuge tests are presented in the prototypescale.) The selected physical models have various configurations of soil layers (H NL& H L ) and distance between foundation and quay wall (s). Test numbers 1 to 17 arethe tests on single pile and the last test number 18 is on a pile group. The pile foundationin No. 18 consists of four piles rigidly fixed to the footing having a pile spacingof 2.5D. All the piles used in the physical model tests do not reach their elasticlimits and behave as elastic beams.NUMERICAL PROCEDUREThe beam on non-linear Winkler foundation method with hyperbolic type p-ycurves is used. The vertical beam models the pile and the horizontal springs connect-2

ing the beam and supporting ground model soils. The governing differential equationfor the model can be expressed as4d yEI4dz= −Dpand the p-y springs modelled by a hyperbolic function can be written askhrp = Ci1+yyrywhere EI=flexural rigidity of pile, y=relative displacement between pile (u) and soilsin free field (u g ), z=depth from the pile head, D=width of pile (or width of footing, B),p=horizontal subgrade reaction, C i =scaling factor for the p-y curve at i-th layer,k hr =coefficient of initial subgrade reaction parameter, and y r =reference relative displacement.As p = ky hry= ∞ rwhen C i =1, y r may be defined as follow using theBroms’s ultimate pile resistance (1964):y = 3 K σ ′rPkhrwhere K P =coefficient of passive earth pressure, and σ v′ =effective overburden pressure.In order to systematically determine the soil parameters for various test conditions,the following assumptions are made:・ φ ′ for all the sands used is assumed as 40 degrees, as materials used in the physicalmodel tests are mainly clean sand. As a result, K P =4.6 for all the cases.・ Shear modulus of sands is assumed to be expressed as G ()( ) 0. 400= 75F e σ ′vpawhere G 0 =shear modulus at very small strain (in MPa), e=void ratio,2F () e = ( 2.17 − e) ( 1 + e), and p a =atmosphere pressure. This relationship is obtainedfrom bender element tests on Toyoura sand in a triaxial cell (Takahashi,E = 2 + ν G .2004). By assuming ν=0.3, Young’s modulus,0( 1 )0・ k hr is estimated by (JRA, 2002):khr−34Es⎛ BE⎞= αi ⎜ ⎟ = αik0.3 ⎝ 0.3⎠h0where α i =additional scaling factor for the coefficient of initial subgrade reactionparameter at i-th layer, E s =soil’s Young’s modulus, and B E =effective width offoundation (=B (width of foundation) for footing, and = D β for pile where β=stiffness ratio of soil to pile (= 4 k h 0D4EI , in 1/m)). k0for β is the averageh3

value of k h0 from the depth of pile head to 1/β. Since it is common to choose k h at10 -2 m order of displacement (strain level of 10 -2 ) as the coefficients of the initialsubgrade reaction in the practical pile design procedure, E s is assumed to beE 0 /10.Since the analyses presented here are performed at a snapshot in time, selection ofthe target time in an earthquake is crucial. In this study, the time just after shaking isselected and the measured soil displacement profile at the time is input, since (1) ourmain concern in this paper is assessment of p-y curve parameters for predicting permanentfoundation deformation due to liquefaction-induced lateral spreading of soils,and (2) the horizontal pile displacement monotonically increases with shaking andceases to increase at the end of shaking for all the physical model tests. (No superstructureinertial force is considered.) The other boundary conditions such as constraintconditions of the beam-ends corresponding to the pile head and pile tip are setaccordingly.dpdyIn the above equations, C i and α i are fitting parameters: asy=0= α C kiih0p σ ′max= C i3K P vand, C i can be used for scaling of overall p-y curve and α i can give additionalreduction to the initial subgrade reaction. These may be governed by magnitudeof accumulated excess pore water pressure for the liquefiable layer and probablyloosening of soil for the surface non-liquefiable layer as well as effects of accumulationof displacements in cyclic loading in the monotonic loading analysis.C i and α i are estimated by minimising following residual error (∆E):∆E=N∑i=1*⎛ Mi− M⎜⎝ Mii2⎞⎟ wi⎠*where N=number of bending moment measurement points, M i =estimated bendingmoment at i-th measurement point, Mi=measured bending moment, and22w = M M (weighting factor).ii∑ANALYSIS RESULTSiIn the first trial (Case 1), estimation of C i is made with α i =1. From now onward, parametersof i=L represents those for the liquefiable layer and i=NL is for the surfacenon-liquefiable layer. Estimated combinations of scaling factors (C * i) satisfying∆E

oth the non-liquefiable and liquefiable layers (µ NL (UCL) & µ L (UCL) ) , et cetera. M maxcan be reasonably predicted as the mean scaling factors do when the lower confidencelimit value is not used for the liquefiable layer. These calculations indicate that(1) the parameter change for the surface non-liquefiable layer is less sensitive to thepile response than that for the liquefiable layer, and (2) conservative prediction of themaximum bending moment in a pile can be reasonably made with the scaling factorgreater than the expected mean value determined for the liquefiable layer and thosewithin a range of its confidence limits determined for the surface layer when only thekinematic loadings due to large horizontal movement of soils are considered.In the calculations made in this study, only the kinematic loadings due to large horizontalmovement of soils are considered and conservative prediction of maximumbending moment can be made using stiffer/stronger p-y curves. However, it may benot the case when the superstructure inertial forces are also considered. Expectedmean values for the scaling factors would be appropriate for the pile response assessmentswhen both the kinematic and inertial loadings are considered.Further examinations for piles in laterally spreading soils having various boundaryconditions, e.g., piles located behind different types of quay wall, those subjected tohorizontal movement of soils having various displacement profiles, etc., may beneeded to expand scope of application.CONCLUSIONSVariation of p-y curve parameters for piles in laterally spreading liquefied soils issystematically examined by numerically analysing physical model tests undertakenby independent researchers using the beam on non-linear Winkler foundation methodwith hyperbolic type p-y curves. Estimated best-fit parameters for p-y curve arewidely scattered as expected. Key findings obtained are; (1) the parameter change forthe surface non-liquefiable layer is less sensitive to the pile response than that for theliquefiable layer, and (2) conservative prediction of the maximum bending moment ina pile can be reasonably made with the scaling factor greater than the expected meanvalue determined for the liquefiable layer and those within a range of its confidencelimits determined for the surface layer when only the kinematic loadings due to largehorizontal movement of soils are considered. Further examinations for piles in laterallyspreading soils having various boundary conditions may be needed to expandscope of application.ACKNOWLEDGEMENTThe authors are grateful to Prof S. Yasuda and Dr T. Tanaka of Tokyo Denki Universityfor allowing them to use unpublished physical model tests data.REFERENCES6

Brandenberg, S.J., Singh, P., Boulanger, R.W., and Kutter, B.L. (2001). “Behavior ofpiles in laterally spreading ground during earthquakes.”, Proc. 6 th Caltrans SeismicResearch Workshop, CA, Paper 02-106.Broms, B.B. (1964). “Lateral resistance of piles in cohesionless soils.” J. Soil Mech.Found., ASCE, Vol.90, No.SM3, 123-156.Fujiwara, T., Horikoshi, K. and Sueoka, T. (1997). “Centrifuge model tests on gravitytype quay wall.” Proc. Sym. Liquefaction-induced lateral Spreading of Soils, Tokyo,Japan, 235-240 (in Japanese).Fujiwara, T., Horikoshi, K. and Sueoka, T. (1998). “Dynamic behavior of gravitytype quay wall and surrounding soil during earthquake.” Proc. Centrifuge 98, Tokyo,Japan, 359-364.Horikoshi, K., Tateishi, A. and Fujiwara, T. (1998). “Centrifuge modeling of a singlepile subjected to liquefaction-induced lateral spreading.” Special Issue of Soils.Found., No.2, 193-208.Japan Road Association (2002). Specification for Highway Bridges, Part IV, JRA,Tokyo, Japan.Japan Society of Civil Engineers (2000). Earthquake Resistant Design Codes in Japan,JSCE, Tokyo, Japan.Public Works Research Institute (1998a). “Shaking table tests on piles subjected toliquefaction-induced laterally spreading soils.” Internal Report, PWRI, Tsukuba,Japan (in Japanese).Public Works Research Institute (1999b). “Dynamic centrifuge model tests on liquefactionremediation for mitigation of damage of piles subjected to liquefactioninducedlaterally spreading soils.” Internal Report, PWRI, Tsukuba, Japan (inJapanese).Sento, N, Yanagisawa, E., and Fujiki, H. (1998). “1g shaking table tests on differentflexural rigidity piles in liquefaction-induced lateral spreading of soils.” Proc.33 rd Japan National Conf. Geotech. Eng., Yamaguchi, Japan, 1003-1004 (inJapanese).Takahashi, A. (2004). “TC29 bender element round robin test results.” Soil MechanicsSection Internal Report, Imperial College London, UK.Yasuda, S., Tanaka, T. and Ishii, T. (2004). “Adaptability of pile installation methodas a countermeasure against liquefaction-induced flow.” Proc. 15 th SoutheastAsian Geotech. Conf., Bangkok, Thailand, 917-922.Yasuda, S. and Tanaka, T. (2005). Personal communication.7

No. CASoilsSurface layerTable 1: Summary of physical model testsH NLLiquefiable layerH LBase layerInput motion1 1LooseToyoura sand Toyoura sand Sinusoidal waves0.100.30fine gravel(Dr=50%)(H=0.25m, Dr=65%) (0.15G, 10Hz, n=20)2 1LooseToyoura sand Toyoura sand Sinusoidal waves0.100.30fine gravel(Dr=50%)(H=0.25m, Dr=65%) (0.15G, 10Hz, n=20)3 1LooseNikko silicaNikko silica sand Sinusoidal waves0.100.40fine gravelsand (Dr=50%) (H=0.27m, Dr=50%) (0.40G, 3Hz, n=20)4 1LooseNikko silicaNikko silica sand Sinusoidal waves0.000.50fine gravelsand (Dr=50%) (H=0.27m, Dr=50%) (0.40G, 3Hz, n=20)5 1LooseNikko silicaNikko silica sand Sinusoidal waves0.100.40fine gravelsand (Dr=50%) (H=0.27m, Dr=50%) (0.40G, 3Hz, n=20)6 1LooseNikko silicaNikko silica sand Sinusoidal waves0.100.40fine gravelsand (Dr=50%) (H=0.27m, Dr=50%) (0.40G, 8Hz, n=20)7 1Toyoura sand Toyoura sand Toyoura sand Sinusoidal waves0.501.00(Dr=50%)(Dr=50%)(H=0.3m, Dr=85%) (0.50G, 5Hz, n=20)8 50Toyoura sand Toyoura sandSinusoidal waves0.007.25 --(Dr=55%)(Dr=55%)(0.15G, 1Hz, n=60)9 50Toyoura sand Toyoura sandSinusoidal waves0.007.25 --(Dr=55%)(Dr=55%)(0.15G, 1Hz, n=60)10 50Toyoura sand Toyoura sandSinusoidal waves0.007.25 --(Dr=55%)(Dr=55%)(0.15G, 1Hz, n=60)11 50Toyoura sand Toyoura sandSinusoidal waves3.254.00 --(Dr=55%)(Dr=55%)(0.15G, 1Hz, n=60)12 50Toyoura sand Toyoura sandSinusoidal waves3.254.00 --(Dr=55%)(Dr=55%)(0.15G, 1Hz, n=60)13 50Toyoura sand Toyoura sandSinusoidal waves1.505.75 --(Dr=55%)(Dr=55%)(0.15G, 1Hz, n=60)14 50Toyoura sand Toyoura sandPort Island-NS1.505.75 --(Dr=55%)(Dr=55%)@GL-16.4m (0.26G)15 50Toyoura sand Toyoura sandPort Island-NS1.754.25 --(Dr=40%)(Dr=40%)@GL-16.4m (0.26G)16 50Toyoura sand Toyoura sandSinusoidal waves1.754.25 --(Dr=40%)(Dr=40%)(0.19G, 1Hz, n=20)17 50Toyoura sand Toyoura sandSinusoidal waves1.754.25 --(Dr=40%)(Dr=40%)(0.10G, 1Hz, n=20)18 * 50Silica sandSilica sandSilica sand No.7 Sinusoidal waves3.007.00No.7 (Dr=60%) No.7 (Dr=60%) (H=10m, Dr=90%) (0.24G, 1.5Hz, n=24)CA : Centrifugal acceleration (G)H NL : Thickness of surface non-liquefiable layer (m)H L : Thickness of liquefiable layer (m)* Foundation consists of 4 piles rigidly fixed to the footing (Pile spacing=2.5D)8

No.Table 1: Summary of physical model tests (continued)Foundation Test results p-y model † p-y model ‡D EI βL s/H u q /H u g /D u/D C * NL C * L α * NL C * LSource1 0.018 1.23E-5 7.33 0.88 --Sento et al.0.94 1.28 0.00 0.145 0.300 0.100(1998, Small EI)2 0.025 9.22E-4 2.60 0.88 --Sento et al.0.72 0.07 0.65 0.070 0.026 0.070(1998, Large EI)3 0.030 2.13E-5 7.98 0.80 0.26 ** Yasuda et al.4.37 4.13 0.00 0.180 0.000 0.185(2004, Case 1)4 0.030 2.13E-5 9.06 0.80 0.29 ** 8.27 1.33 -- 0.060 --Yasuda & Tanaka0.060(2005., Case 2)5 0.030 1.70E-4 4.85 0.80 0.32 ** Yasuda & Tanaka5.13 1.77 1.00 0.325 0.670 0.325(2005, Case 3)6Yasuda & Tanaka0.030 2.13E-5 7.98 0.80 0.14 1.47 1.83 0.00 0.120 0.000 0.120(2005, Case 4)7 0.060 5.54E-2 3.01 0.33 0.19 4.58 -- 1.00 0.075 0.670 0.085 PWRI(1998a, Case 6)8 0.325 8.26E+0 5.11 0.28 0.14 3.48 0.83 -- 0.090 --Horikoshi et al.0.090(1998, Case A)9 0.325 8.26E+0 5.11 1.31 0.12 2.68 0.89 -- 0.095 --Horikoshi et al.0.095(1998, Case B)10 0.325 8.26E+0 5.11 2.00 0.10 2.14 0.38 -- 0.075 --Horikoshi et al.0.075(1998, Case C)11Horikoshi et al.0.325 8.26E+0 4.93 0.28 0.09 1.74 1.46 0.10 0.040 0.003 0.055(1998, Case E)12Horikoshi et al.0.325 8.26E+0 4.93 2.00 0.09 1.20 1.51 0.00 0.245 0.000 0.260(1998, Case F)13Horikoshi et al.0.325 8.26E+0 4.58 0.28 0.10 2.14 0.55 0.15 0.030 0.002 0.025(1998, Case G)14Horikoshi et al.0.325 8.26E+0 4.58 0.28 0.04 0.63 0.48 0.00 0.070 0.002 0.060(1998, Case H)15Fujiwara et al.0.400 5.25E+1 2.41 1.88 0.18 0.99 0.15 0.15 0.180 0.005 0.185(1998, Case A)16Overturned(1997, Case B')Fujiwara et al.0.400 5.25E+1 2.41 1.88 1.90 0.20 0.25 0.135 0.003 0.15017Fujiwara et al.0.400 5.25E+1 2.41 1.88 0.23 0.45 0.08 0.15 0.050 0.008 0.070(1998, Case B)18 * 0.500 3.70E+2 5.08 0.98 0.13 2.00 0.56 0.40 0.005 0.012 0.005 PWRI(1998b, Case 1)D : Pile diameter (m)EI : Flexural rigidity of pile (MN.m 2 )βL : Dimensionless length of pile (β=stiffness ratio of soil to pile & L=pile length)s : Distance of foundation from quay wall (m)H : H NL + H L (m)u q : Horizontal displacement of top of quay wall (m)u g : Horizontal displacement of ground surface at foundation location (m)u : Horizontal displacement of foundation (m)C * NL : Estimated scaling factor for non-liquefiable layer in p-y modelC * L : Estimated scaling factor for liquefiable layer in p-y modelα * ΝL : Estimated scaling factor for k hr for non-liquefiable layer** Heal settlement of quay wall is greater than that at the toe.† Best-fit parameters with α NL =1 and α L =1 (Case 1).‡ Best-fit parameters with C NL =1 and α L =1 (Case 2).9

Table 2: Summary of expected parameters for p-y curvesCase 1 Case 2C NL α NL C L α L C NL α NL C L α L(mean)µ i 3.0E-1 1 8.2E-2 1 1 1.9E-2 8.3E-2 1LND(UCL)µ i 5.9E-1 -- 1.3E-1 -- -- 8.3E-2 1.3E-1 --(LCL)µ i 1.6E-1 -- 5.1E-2 -- -- 4.2E-3 5.2E-2 --(mean)µ i 2.8E-1 1 1.1E-1 1 1 1.2E-1 1.1E-1 1ND(UCL)µ i 4.8E-1 -- 1.5E-1 -- -- 2.6E-1 1.5E-1 --(LCL)µ i 7.0E-2 -- 7.1E-2 -- -- (-1.9E-2) 7.1E-2 --LND: Logarithm of parameters is assumed to be normally distributed.ND: Parameters are assumed to be normally distributed.µ (mean) i : Mean value of parameterµ (UCL) i : Upper confidence limit with risk degree of 0.05 (confidence coefficient of 0.95)µ (LCL) i : Lower confidence limit with risk degree of 0.05 (confidence coefficient of 0.95)Lateral spreading of soilsSoil movementimpositionPileSurface layerLiquefiablelayerBase layerQuaywallWinkler foundation<strong>Beam</strong> modelling pileTarget pile foundation in laterally spreading soilsSchematic view of analytical model usedFigure 1: Schematic view of problem to be solvedScaling factor for p-y curvein liquefiable layer, C* L0.50.40.30.20.10No.01 No.02 No.03 No.05No.06 No.07 No.11 No.12No.13 No.14 No.15 No.16No.17 No.18C L =C NLResidual error:∆ E < 0.010 0.2 0.4 0.6 0.8 1Scaling factor for p-y curve in non-liquefiable layer, C* NLFigure 2: Combinations of scaling factors for p-y curve whose residual error,∆E

1No.7No.5Scaling factor for p-y curve0.80.60.40.2No.2Estimated scaling factors:C* NLC* L(Filled markers for centrifuge tests.)00 2 4 6Average normalised relative displacement between pile and soil in free field, | y average |/DFigure 3: Best-fit p-y curve scaling factors (C * NL & C * L) change with average relativedisplacement between pile and free-field soil (Case 1)Scaling factor for p-y curve10 010 -110 -210 -310 -410 -510 -610 -7No.1No.6No.12No.3No.7No.5Estimated scaling factors:α* NL with C NL =1C* L(Filled markers for centrifuge tests.)0 2 4 6Average normalised relative displacement between pile and soil in free field, | y average |/DFigure 4: Best-fit p-y curve scaling factors (α ∗ NL & C * L) change with average relativedisplacement between pile and free-field soil (Case 2)5Error in maximum bending moment (%)400Use of stiffer/strongersoil springs 300Use of softer/weakersoil spring(mean)200µ NL (LND)(mean)µ L (LND)(mean)Overestimationµ NL (ND)100(mean)µ L (ND)No.18180Under--estimation No.3 0 7 27No.735 5-100No.5Difference between best-fit and mean scaling factor, ( C* NL −µ NL (mean) )/σ NL or (C* L −µ L (mean) )/σ LFigure 5: Errors in maximum bending moment with mean scaling factors (Case 1)11

Error in maximum bending moment (%)No.181813400No.133002001000-2 0 2-1005(mean)µ NL(mean)µ L(mean)µ NL(mean)µ LNo.5(LND)(LND)(ND)(ND)Difference between best-fit and mean scaling factor, ( α* NL −µ NL (mean) )/σ NL or (C* L −µ L (mean) )/σ LFigure 6: Errors in maximum bending moment with mean scaling factors (Case 2)400300µ NL(UCL)µ L(UCL)400300µ NL(UCL)µ L(LCL)200200Error in maximum bending moment (%)No.1810000-4 -2 No.7 0 7 2 4 -4 -2 0 2 4-100400300200No.5µ NL(LCL)µ L(UCL)100-100400300200µ NL(LCL)µ L(LCL)No.181000-4 -2 0 2 4No.77-100 No.5 5100-4 -200 2 4-100Difference between best-fit and expected scaling factor, ( C* NL −µ NL )/σ NL or (C* L −µ L )/σ LFigure 7: Errors in maximum bending moment with expected scaling factors (Case 1,Logarithm of parameters is assumed to be normally distributed.)12