A CENTRIFUGE MODEL STUDY ON THE EFFECTSOF PILE INSTALLATION PROCESS ON SEISMIC

BEHAVIOR OF PILED RAFT FOUNDATIONFOR OIL STORAGE TANKS

Seyed Mohammad Sadegh SAHRAEIAN1, Jiro TAKEMURA2 and Sakae SEKI3

1Post-Doctoral Research Fellow, Dept. of Civil Eng., Shiraz University;Former Ph.D. Candidate, Dept. of Civil Eng., Tokyo Institute of Technology

(Civil Engineering Department, School of Engineering, Zand Street, Shiraz 7134851154, Iran)E-mail: sadegh.sahraeian@shirazu.ac.ir

2Member of JSCE, Associate Professor, Dept. of Civil Eng., Tokyo Institute of Technology(M5-205, 2-12-1, Ookayama, Meguro-ku, Tokyo 152-8552, Japan)

E-mail: jtakemur@cv.titech.ac.jp3Technician, Dept. of Civil Eng., Tokyo Institute of Technology

(M5-205, 2-12-1, Ookayama, Meguro-ku, Tokyo 152-8552, Japan)E-mail: seki@cv.titech.ac.jp

Some level of settlement is allowed in the design of oil tanks if uneven settlement is controlled withinallowable values. Considering the critical condition of piled raft foundation (PRF), that is, secure contact ofraft base to the ground surface, PRF is considered one of the rational foundations for the oil tanks. However,PRF has a complex interaction with soil under horizontal seismic loading, especially if the tank rests on aliquefiable soil. On the other hand, the pile installation method can affect the pile bearing capacity and theliquefaction resistance of sand as well. In this study, a series of centrifuge tests was performed to investigatethe mechanical behavior of oil tanks supported by PRF on liquefiable sand. In the tests, slab and piled-raftfoundations were modeled. In the case of PRF, two different methods of pile installation (Driven and Non-Driven) were modeled and the Driven PRF models were made with two different pile numbers. Using theobserved results, such as accelerations of the tank and ground, displacements of the foundation and excesspore water pressures of the ground, advantages, and limitations of PRF for oil tanks on liquefiable sand arediscussed.

Key Words: oil storage tank, liquefaction, piled raft foundation, centrifuge modeling, driven and non-driven piles, piles number

1. INTRODUCTION

The majority of existing oil storage tanks in Japanwere constructed before the early 1970s when soilliquefaction was first considered in the design oftank foundations. After the 1964 Niigata earthquake,the 1978 Miyagiken-oki earthquake and the 1995Hyogoken-Nambu (Kobe) earthquake, it has becomean urgent matter for geotechnical engineers to assessthe seismic stability of existing oil storage tanks andimplement proper countermeasures against soil lique-faction.1), 2)

Since the concept of piles as settlement reducerswas introduced by Burland et al. (1977),3) piled raftfoundation (PRF) has received remarkable attention,especially in reducing the construction expenses byreducing the required number of piles. The raft inthis foundation system has adequate bearing capac-

ity; therefore, the main objective of introducing thepile elements is to control or minimize the settle-ment, especially uneven settlement, rather than tocarry the major portion of the vertical loads. There-fore, a major question on this type of foundation ishow to design the piles optimally to control the set-tlement.4), 5) This foundation has been consideredfor building foundation design and some case stud-ies have been reported.6) Yamashita et al. (2014)7)

investigated the performance of PRF during the 2011off the Pacific coast of Tohoku earthquake. Besides,some researchers have observed the complicated be-havior of PRF under other loading conditions. Past-sakorn et al. (2002)8) studied the behavior of thisfoundation system under static lateral loading tests in1 g condition to evaluate the application of pile groupsand PRF and discuss the optimized parameters, e.g.,raft size, number of piles and piles spacing. In order

Journal of JSCE, Vol. 5, 357-376, 2017

357

to study the performance of PRF in dynamic loadingcondition, some researches were also accomplished.9)

In addition, centrifuge modeling has also been usedunder static and dynamic loadings to study the me-chanical behavior of PRF. A comparison of piled raft,pile group, and raft foundations under static hori-zontal loadings were made using centrifuge modeltests.10), 11) On the other hand, dynamic centrifugemodel tests were also conducted by Horikoshi et al.(2003)12) to compare a piled raft with a pile groupfoundation and discuss the effect of pile head connec-tion to the raft. In spite of enormous studies on PRFs,optimal and rational design methods of this founda-tion have not been extended to civil engineering in-frastructures. This is partly due to the complex soil-structure interaction between raft-ground-piles duringan earthquake. In particular, if the piled raft rests on aliquefiable ground, the soil-foundation interaction be-comes more complicated. Due to this complexity andpossible large settlement, the practical implementa-tion of PRF is not a straightforward issue.

Securing the contact of the raft to the subsoil isanother concern in the seismic design of PRF. Thecontribution of raft against the horizontal loads can-not be assured without this contact. For this purpose,the raft settlement should be equal to or greater thanthat of the surrounding ground; otherwise, a gap be-tween the raft and the ground surface may develop.In the design of oil tank foundation, the main concernregarding settlement is uneven settlement, not maxi-mum settlement. For example, an allowable unevensettlement is 1/300 of tank diameter,13) which impliesthat some level of tank foundation settlement is per-mitted if the uneven settlement is controlled below theallowable value. Thus, this foundation system is con-sidered one of the rational foundations for oil storagetanks. Some studies have been done on oil tank foun-dations. For example, performances of pile founda-tion of storage tanks were investigated in some casestudies.2), 14) Also, some case studies were reportedby Ishihara, et al. (1980)1) and Sento et al. (2004)15)

about oil tanks on liquefiable sandy soil and soil im-provement methods as the countermeasure. A few re-searchers have considered PRF for the storage tanks.Liew et al. (2002)16) reported a case study of oil stor-age tank with PRF. A finite element method (FEM)was utilized by Chaudhary (2007)17) to study the be-havior of PRF for a huge storage tank. Also, De Sanc-tis and Russo (2008)18) reported a PRF design for acluster of sodium hydroxide circular tanks. Further-more, among the studies on PRF of oil tanks on lique-fiable sand, Imamura et al. (2010)19) and Takemura etal. (2014)20) utilized centrifuge modeling to investi-gate the dynamic response of an oil tank supported byPRF. Although dynamic and permanent settlementsand rotations of foundations were observed in theseresearches, the observations were only made in the

shaking direction, not in the different directions.On the other hand, the bearing capacity of pile

foundation depends on the pile installation proceduresince it changes not only the structural behavior of thepiles but also the stiffness and strength of the soil. Inorder to investigate this issue, some researches havebeen conducted on the pile installation effect on theload capacity of piles. Adejumo and Boiko (2013)21)

utilized field tests on driven and non-driven (bored)piles in layered sandy clay soil to study the effectof the installation method. Also, the installation andload testing of some driven cast-in-situ (DCIS) pilesat a uniform sand site were reported by Flynn andMcCabe (2015).22) Besides these studies, some re-searchers conducted centrifuge model tests to exam-ine the effect of the pile installation procedure on pilebehavior. Bloomquist et al. (1991)23) developed a piledrive-load test device for in-flight installation of pilesin sand and compared the pile load capacity of 1 gstatically pushed piles and in-flight driven piles. An-other in-flight pile driver with loading set was man-ufactured by Pan et al. (1999)24) which could drivepiles, one by one into the soil. Furthermore, a se-ries of model pile tests was conducted by centrifugeto study driven piles behavior installed by a miniaturepile driving actuator in homogeneous silica sand withvarious densities.25) Also, a comparison between theseismic behavior of driven and non-driven PRF of oiltanks was presented by Sahraeian et al. (2015)26) us-ing centrifuge modeling. Despite these previous stud-ies, the difference between the behavior of PRF withdriven and non-driven piles during dynamic loadinghas not been well investigated. On the other hand, al-though there are some studies about PRF in the litera-ture, the effect of piles number on the piled raft foun-dation of oil tanks on liquefiable sand has not beenconsidered.

In this study, in order to investigate the mechan-ical behavior of oil tank supported by PRFs withdriven and non-driven piles on the liquefiable satu-rated sand, dynamic centrifuge model tests were per-formed. For this objective, a special setting was de-veloped to model a driven PRF for the oil tank in-flight. Moreover, for the driven piles PRF, the ef-fect of piles number on the performance of PRF wasalso investigated. From the observed test results, suchas excess pore water pressures and accelerations ofthe ground, and accelerations, rotation and settlementof the tank, typical dynamic and permanent displace-ments of the tank with PRF are studied in comparisonto those of the slab foundation. In particular, the be-havior of tank was observed not only in the shakingdirection but also in the transverse direction. Fromthese investigations and comparisons, the advantagesand limitations of PRF with driven and non-drivenpiles in the application to the oil tanks on liquefiablesandy soil are discussed.

358

Fig.1 Model setup, instrumentation and laminar box used for the tests.

Table 1 Test cases.

Testcode

Foundation Ground Details Malfunctioningsensors

Case 3b Slab Saturated sand (Dr = 68%) 5 built-in EPs P2, P3Case 4 Non-Driven Piled Raft (12 piles) Saturated sand (Dr = 69%) 5 non-built-in EPs A4, EP5Case 5 Driven Piled Raft (12 piles) Saturated sand (Dr = 70%) 5 built-in EPs P4, P11, P12Case 6 Driven Piled Raft (24 piles) Saturated sand (Dr = 65%) 5 built-in EPs P4, P5, P8, A11, A13

2. DYNAMIC CENTRIFUGE TESTS

(1) Equipment, model foundations, and testcases

Using Tokyo Tech Mark III centrifuge and a shak-ing table,27) centrifuge tests were conducted under50 g centrifugal acceleration. A laminar box con-sisted of 15 laminas and rubber membrane bag withinner dimensions of 600 mm in length, 250 mm inwidth, and 438 mm in depth was used for making themodel setups shown in Fig. 1.

Four model tests were performed as shown in Ta-ble 1. In Case 3b, a slab foundation (SF) was placedon the saturated sand. A non-driven piled raft founda-tion (PRF) including 12 piles was placed on saturatedsand in Case 4. To compare the behavior of oil tankssupported by PRFs with non-driven and driven piles,a driven piled raft foundation with 12 piles was mod-eled on saturated sand in Case 5. Based on a founda-tion design method, e.g., Meyerhof’s method28) andusing the φ′ of Silica sand No.8 (Table 2), the bear-ing capacity of each pile and raft in the prototypescale are about 1.2 MN and 107 MN, respectively inthe static condition. Namely, the bearing capacity of12 piles is about 14% of the bearing capacity of theraft. In order to increase the similarity of this foun-dation model with the real oil tank foundations interms of piles number, the driven piled raft founda-tion with larger piles number (24 piles) was modeledin Case 6. For non-driven piles, PRF with 24 piles

Table 2 Properties of silica sands.

No.8 No.3

Specific gravity Gs 2.65 2.56Mean grain size D50 (mm) 0.1 1.47Effective grain size D10 (mm) 0.041 1.21Uniformity coefficient Uc 2.93 1.26Maximum void ratio emax 1.333 0.971Minimum void ratio emin 0.703 0.702Effective friction angle φ′ (◦) 41Permeability coefficient k (m/s) 2.0 × 10−5 4.6 × 10−3

(prototype for 50 g) (1.0 × 10−3) (2.3 × 10−1)

was not modeled in this study due to the difficulties inpreparing the sand using the method employed (seesection 3. a)) for the small pile spacing. During theground preparation, the sensors were placed in twodifferent sections, the first section the center line ofthe model in the shaking direction and the second inthe transverse direction. It should be noted that the ac-celerations in the transverse direction were measuredonly at the bottom (A12) and top (A13) of the tankmodel, and the ground acceleration in the transversesection (A10 and A11) are in the shaking direction(Fig. 1(c)). Model dimension and instrumentation de-tails are shown in Fig. 1(a), 1(b), and 1(c).

(2) Tank, pile, raft, and ground modelingTable 3 shows the characteristics of the tank, pile,

359

Table 3 Characteristics of tank, raft, and pile used inmodel and prototype in 50 g.

Model Prototype

Tank

material acrylic cylinder steelouter diameter 140 mm 7.0 mthickness 3 mmheight 160 mm 8.0 mmass (liquid and raft) 2.9 kg 363 tvertical load intensity 81 kPa 81 kPa

Raft

material aluminum RCdiameter 150 mm 7.5 mthickness 10 mm 0.5 mbase surface rough rough

Pile

material aluminum RCouter diameter 6 mm (0.5 mm) 0.3 mthickness 0.5 mm 25 mmaxial rigidity: EA 596 kN 1.49 GNbending rigidity: EI 0.0023 kNm2 14.2 MNm2

shaft surface rough rough

Fig.2 Tank and raft models.

and raft for both the model and prototype scales. Thetank model (Fig. 2(a)) is made of an acrylic cylin-der with 140 mm outer diameter, 160 mm height, and3 mm thickness. It was glued to the slab/raft modelmade of an aluminum disk with a diameter of 150 mmand thickness of 10 mm (Figs. 2(a) and 2(c)). The raftmodel has 12 (Cases 4 and 5) and 24 (Case 6) coni-cal shape concave holes, which are put on the pileheads (Figs. 2(d), 2(e), and 3). Silica sand No. 8 wasglued on the bottom surface of the raft model to cre-ate a rough surface condition. Water was used as aliquid inside the tank with a height of 140 mm. Thetotal weight of the water, tank and raft (2.9 kg), in-

Fig.3 Pile model.

duced 81 kPa of the vertical load intensity (qv: totalweight/raft base area) under 50 g centrifugal acceler-ation.

The piled raft foundation had 12 or 24 identicalpiles, made of an aluminum tube with outer diam-eter of 6 mm, a thickness of 0.5 mm, and length of100 mm as shown in Fig. 3. These piles were ar-ranged symmetrically (Figs. 2(d) and 2(e)). The pileheads were not rigidly fixed to the raft, but simplycapped by the concave hole, which allows partiallyfree rotation like pinned connection (Fig. 3). In thisway, the piles mostly were subjected to large axialand lateral forces and a small bending moment at theconnection point to the raft. This condition is closeto the actual situation of the normal pile foundationof oil tanks.29) Rough pile shaft surface was alsomade by gluing silica sand No.8 (Fig. 3). The con-tact pressures at the raft base were measured by fiveearth pressure (EP) cells. In Case 4, external (non-built-in) cells were glued on the raft base, while inCase 3b, Case 5 and Case 6, new raft models with fivebuilt-in earth-pressure cells covered by thin siliconrubber were employed to improve the reliability ofearth pressure measurements by eliminating the stressconcentration on the attached EP cells (Fig. 2(b)). Al-though the type of sensors was different in Case 4, asshown later, the measured data were acceptable fordiscussing the general trend of load sharing betweenthe piles and raft. For the estimation of vertical loadproportion between the piles and raft, it is also possi-ble to measure the pile axial load using instrumentedpiles.20) However, in the pile installation process ofthe driven piles, the instrumented piles with wirescannot be used. Therefore, the average measured raftcontact pressures by EP cells were employed to esti-mate the load sharing between the piles and raft.

360

Fig.4 Guide and pilesduring sand pouring(Case 4).

Fig.5 Exerted loads inpreloading process(Cases 3b & 4).

(3) Model preparations and test proceduresa) Sand preparation

Fine silica sand (No. 8) was used for the liquefiablesand layer and coarse silica sand (No.3) for the bot-tom drainage layer. Detailed properties of silica sandsNo. 3 and No. 8 are presented in Table 2. Water wasused as pour fluid of the sand. The prototype perme-ability of silica sand No.8 is about 1.0 × 10−3 m/s in50 g centrifugal acceleration. Although this value isrelatively high, it is low enough to accumulate excesspore water pressure and create complete liquefactionwith water in the short duration of input motion used.The coarse grain size silica sand No. 3 was utilizedas the drainage layer at the model bottom to supplywater evenly into the model ground during the satu-ration process. Using the air pluviation method, thesand layer was made with a target relative density of65%. But in some cases, the final relative density hada few deviations from the target value (Table 1).

In the case of non-driven PRF (Case 4), the pileswere fixed at the center of the model box by an alu-minum guide (Fig. 4). Then, using the air pluviationmethod, sand was poured until reaching the requiredlevel. At this level, the piles tip with conical shapewas just above the ground surface and could be putinto the concave holes at the raft base. During thesand preparation, the accelerometers and pore waterpressure transducers (PPTs) were placed at the pre-scribed locations as shown in Fig. 1. After makingthe model ground, it was saturated in a vacuum tankby introducing de-aired water from the bottom of themodel box.b) Preloading in Cases 3b and 4

In Cases 3b and 4 after completion of the modelground preparation, the model tank was placed onthe ground. There was inevitable unevenness at theground surface especially for the case with piles,which created non-uniform contact condition of raftbase to the ground surface, such as local gaps. Toreduce the effects of the local bedding error and se-cure the contact, small vertical displacement was im-posed by an electrical jack in 1 g condition. The load-displacement curves measured in the preloadings for

Fig.6 Raft base pressures during preloading (Cases 3b &4).

Cases 3b and 4 are presented in Fig. 5. The load-displacement relations of the two cases were differ-ent due to the different initial bedding conditions andbearing resistances with and without piles. In Case3b, 1 mm displacement was imposed, which exerted245 N of preloading. While in Case 4, 490 N wasapplied assuming the raft load proportion (RLP=raftload/total vertical load) of 50%, which could exert thesame raft load as Case 3b. There was no particularreason for RLP=50%, but it was employed as the firstapproximation. However, the substantial displace-ment exerted to the ground by the raft in Case 4 couldbe almost the same as that of Case 3b, 1 mm, whichcan be estimated from the displacement after the bentof load-displacement curve of Case 4. The bent couldbe an evidence of additional resistance from the raft.The measured earth pressures in Case 3b with thebuilt-in cells and Case 4 with the non-built-in cellsduring the preloading process are presented in Fig. 6together with the vertical load intensity (qv: the loadexerted by the self-weight and the jack divided bythe raft area). In Case 4, the pressure could not berecorded by EP5 (the raft center). In this case, all theearth-pressure cells recorded larger values than themaximum qv (∼25 kPa), which included resistancesnot only from the raft but also from the piles. Theselarger base contact pressures could be attributed to thestress concentration on the non-built-in earth-pressurecells. However, the trends of variation in the mea-sured contact pressures were well comparable to thatof the applied load, meaning that even those nonbuilt-in cells could provide qualitatively useful data duringthe test. However, in order to eliminate this unde-sirable stress concentration and measure a better raftcontribution qualitatively, the built-in earth-pressurecells were implemented at the raft base in Case 3b.Although the measured contact pressures by the built-in cells showed a large difference, the average valueof the measured pressures was much closer to the av-erage exerted pressure (qv). The inevitable unevenground surface condition could be a reason for thelarge variation in the measured contact pressures.

361

c) Pile installation and preloading in Cases 5 and6

In order to model in-flight installation of the drivenpiles in Case 5 (Driven PRF with 12 piles) and Case6 (Driven PRF with 24 piles), a 20 mm thick acrylicguide plate (Fig. 7(a)) was used to hold the pilesand tank during the pile installation at the center ofthe saturated model ground. The plate has verti-cal holes with 6.5 mm diameter at the locations ofpiles. To have upright positions of the piles insertedin the holes, small 10 mm thick Styrofoam pieces with6 mm diameter were used as shown in Figs. 7(a) and7(b). Setting all piles and placing the tank model, thepiles were pushed into the sand in 50 g as the firstpile installation stage using an electrical jack (Fig.7(e)) with a loading rate of 2 mm/min. The jack load-ing was stopped before the tank bottom contacted theguide plate surface. Afterwards, the centrifuge wasstopped, the tank and the guide were removed fromthe top of the piles and then the tank was put againon the top of the piles (Fig. 7(c) and 7(d)). Again,in 50 g centrifugal acceleration, the jack load was ex-

Fig.7 In-flight pile installation process in Cases 5 & 6(Driven PRF).

erted to the tank to drive the piles into the sand modelcompletely and to develop contact between the raftand the sand surface. Furthermore, to have a securecontact of the raft base to the ground surface, preload-ing was applied on the tank in the same in-flight con-dition. The replacement ratios defined by the totalcross-section of the piles to the raft base area are 1.9%and 3.8% for the 12 and 24 piles model, respectively.Assuming that the volume of piles (piles cross area× pile length) is equivalent to the reduction of soilvolume in the pile installation portion (raft base area× pile length), relative density could be increased to70% and 75% for 12 and 24 piles, respectively, fromthe target sand density (Dr = 65%). Due to the staticpile installation procedure and the outward lateral dis-placement of sand at the perimeter of the installa-tion portion, the actual increases in relative densityby the pile driving should be smaller than those val-ues. It should be noted that the model pile installationsequence was different from one-by-one installationin real construction sequence, which is very difficultwith the facility available. However, at least the in-stallation process of driven (displacement) pile couldbe modeled by the procedure employed and the re-sults can be compared with those obtained from thenon-driven (non-displacement) pile model (Case 4).Furthermore, the simultaneous installation of all pilescould avoid uncontrolled unsymmetrical conditionsof sand and piles caused by one-by-one installation.

The load-displacement relationships in the secondinstallation and preloading process in Cases 5 and 6are presented in terms of total load and per pile loadin Figs. 8(a) and 8(b), respectively. The weight of themodel tank and loading block put in the tank is in-cluded in the load. The raft base contacted the sandsurface at the displacement of about 25 mm. Duringthe preloading process, same maximum vertical loadintensity (vertical load/raft area) was applied to thefoundation in the two cases. This load was estimatedby subtracting the load at the time of contact fromthe applied load. Also, the maximum applied preloadwas decided considering the earth-pressure cells ca-pacity (500 kPa). In this way, the maximum applied

Fig.8 Jack loading during 2nd stage of piles installationand preloading (Cases 5 & 6).

362

preload in Case 5 and Case 6 were about 11.5 kNand 13 kN, respectively, and the maximum of verti-cal load intensity (qv) in these cases neglecting pilesload was about 700 kPa in both cases. The per-pileload in Case 6 is almost the same as that in Case 5 atthe end of penetration. However, the load displace-ment curves of the two cases are quite different. Inthe beginning, the per-pile load in Case 5 is largerthan Case 6 and as the penetration increases, the in-crement rate of Case 6 increases while that of Case 5decreases. The larger per-pile load of Case 5 in thebeginning could be attributed to larger sand density(Table 1) and pile group effect, referring to larger pileend bearing capacity in the case of fewer piles. But asthe penetration increased, the larger shaft friction wasmobilized for the latter case than the former due tothe larger soil compaction and the larger horizontalstress. Although the maximum load in the preload-ing process was different in driven-PRF cases (Cases5 and 6) comparing with SF case (Case 3b) and non-driven-PRF case (Case 4), the final penetration depthof raft during the preloading was not so much dif-ferent for all the cases. In Case 3b and Case 4, themaximum settlement during preloading were 1 and2 mm, respectively (Fig. 5), while they were 2.5 mmin Case 5 and 2 mm in Case 6 (Fig. 8). In Fig. 9, raftbase pressures measured by the pressure cells duringthe preloading process are presented with their aver-age raft pressure (ARP) and the vertical load intensity(qv). In spite of some difference in the measured con-tact pressures by built-in pressure cells, the averagevalues of the recorded pressures are nearly compa-rable with the average applied stress with some ex-ception, such as EP3 in the beginning of preloadingof Case 5 and EP4 in the whole loading process ofCase 6. As mentioned before, unavoidable unevenground surface is a reason for significant difference inmeasured pressures especially in the beginning of theloading, which caused small pressures by poor con-tacts and large pressures by stress concentrations. Af-ter completion of the second stage of pile installationand preloading in the driven PRF cases (Cases 5 and

Fig.9 Raft base pressures during preloading (Cases 5 &6).

6), the centrifuge was stopped and the jack was de-tached.d) Shaking tests

After the preloading process, the whole setup wasmounted on the shaking table on the swing platformof the centrifuge. The displacement sensors (LDTs)were set on the model and after filling the tank withwater, the centrifugal acceleration was increased upto 50 g. The shaking tests were conducted after con-firming the steadiness of all sensors output. The targetinput wave of the main shock used in the tests is EWcomponent of the acceleration recorded at Kurikoma,Kurihara city during the 2008 Iwate-Miyagi Nairikuearthquake,30) which is characterized as a vibrationwith a moderate duration. Two shakings were in-putted to the model. Confirming all measured valueswere constant after the first shake, the second shakewith about fifteen percent higher amplitude was in-putted to the model. The comparison of target ac-celeration and its Fourier spectrum with those of in-put motions in the tests are presented in the proto-type scale in Fig. 10. The high-frequency compo-nent of the target motion could not be modeled due tothe limited performance of the hydraulic shaker used.Nevertheless, the Fourier amplitudes were almost thesame for all the test cases for the period longer than0.4 secs, except for Shake 2 in Case 3b. However,there were some differences in the magnitude of in-put acceleration, which can be seen in the variation of

Fig.10 Input accelerations and their Fourier spectra.

363

Fig.11 Arias intensity of input motions.

Arias intensity of the input accelerations in Fig. 11.Arias intensity (Ia) first proposed by Arias (1970)31)

is a measure of intensity of shaking defined as:

Ia =π

2g

∫ ∞0

[a(t)]2 dt (1)

where a(t) is shaking acceleration and t is time.Considering that Ia tends to exaggerate the differencein the acceleration by squaring the acceleration, theinput motions of all cases in Shake 1 are nearly simi-lar, especially until 7 sec. While in Shake 2, remark-ably larger Ia level was obtained in Case 3b than theother cases, which had almost similar input motionlevels mostly until 7 secs. From the time variationof Arias intensity, it can be seen that the majorityof major input acceleration had been exerted until 10to 15 sec and thereafter the input acceleration ampli-tudes were so small and the differences in all caseswere negligible. In the discussion of the test results,the above-mentioned differences in the input motionare taken into consideration.

In the shaking tests, the ground and tank accel-erations, the horizontal and vertical displacementsof the tank, and the excess pore water pressures inthe ground were measured as shown in Fig. 1. De-spite careful instrumentations, some sensors could notmeasure the data, which are shown in Table 1. In thefollowing discussion, the results of the model tests aregiven in the prototype scale unless otherwise stated.

3. RESULTS AND DISCUSSION

(1) Ground responseFourier amplitudes of the ground accelerations

measured in all the cases during Shake 1 are com-pared with that of the input motion in Fig. 12. Al-though there are some missing data due to the mal-functioning of accelerometers and difficulty in plac-ing the sensor in the pile driving portion (Table 1and Fig. 1), several points on the general behavior

and the effects of foundation type can be observedfrom the figures. At the location of A2 (X=Y=0,Z=6.25 m), 1.25 m below the pile ends, the input mo-tions were propagated without attenuation for all thecases. However, above this depth, attenuation of inputmotion is clearly seen in the ground, especially for theshort period components, which is a clear evidenceof soil stiffness reduction by the liquefaction. Theattenuation is more significant beside the tank (A6)than beneath the tank (A5), which can be attributedto the confinement effect of the tank load to the soilunderneath. This could prevent significant reductionof stiffness of the soil under the tank. However, thenatural period of the ground became long due to theliquefaction, which could cause the amplification inthe long period range (over 1 sec) as seen beneath thetank (A5). The acceleration response of A11 locatedat 0.75 m away from the raft edge in the transversesection is similar to that of A5 beneath the tank cen-ter in Cases 3b and 4, implying that the tank loadcould affect the dynamic behavior at the location ofA11. At the location of A6, 3 m away from the raftedge, the attenuation of seismic motion is larger inthe cases of PRFs (Cases 4, 5, and 6) than in the SFcase (Case 3b), implying a more pronounced effect ofthe tank load in the latter case. The same behaviorcan be seen at the location of A3 in Cases 3b and 4.This is because the tank loads were transmitted to thedeep depth through piles of PRFs as compared to SF.Comparing A10 (X=0, Y=4.5 m, Z=3.75 m) in Cases5 and 6, more attenuation can be seen in Case 6 thanin Case 5, which also implies that more tank load wascarried by piles in Case 6 than in Case 5. The detailsof proportion of the vertical load carried by the pilesand the raft will be discussed in Section 3. (3).

(2) Excess pore water pressureThe excess pore water pressures (EPWP) at differ-

ent locations of the ground observed in Shake 1 arepresented in Fig. 13. The figure shows the EPWP indifferent depths in four vertical arrays, three in thelongitudinal section at the tank center (P7, P4, P3,and P2: X=0 m), near (P8 and P5: X=6.75 m) andfar (P9 and P6: X=10 m) outside of the raft, and onein the transverse section (P12, P11: X=0, Y=4.5 m).The EPWPs at P7 in the early stage of shaking is alsoshown in the figure. In Case 3b: P2 and P3; in Case 5:P4, P11, and P12; and in Case 6: P4, P5, and P8 couldnot be recorded (Table 1). In the figure, the initialoverburden stress (σ′v0 = zγ′, γ′: effective unit weightof sand) and the vertical stress (σ′v) are presented. σ′vis the sum of σ′v0 and the stress by the tank pressure,which is calculated by the elastic solution assumingthe uniformly distributed raft pressure on an elastichalf-space.28) Also, excess pore water pressure ratio(ru = EPWP/σ′v) at the same locations of the ground

364

Fig.12 Ground response in shaking direction (Shake 1).

in Shake 1 is shown in Fig. 14. In the early stage ofshaking, the EPWPs increased rapidly and then thisrapid rise ceased and the water pressure either be-came almost constant or increased gradually duringthe shaking. Then the EPWPs started the dissipation.At the locations, outside of the raft (P8, P5, P9 andP6) where the tank load did not affect σ′v, the EPWP

built nearly up to the σ′v value (Fig. 14), meaning al-most zero effective stress (σ′v − EPWP � 0). Whileat the location beneath the tank (P7, P4, P3, and P2)and near the tank (P11 and P12), where the tank loadaffected σ′v, the EPWPs were smaller than the σ′v val-ues, reconfirming the confinement effect of the tankon the soil underneath. The larger the difference be-

365

Fig.13 Excess pore water pressures of the ground in Shake 1.

tween σ′v and σ′v0 is, the larger the remaining effectivestresses, that is, σ′v − EPWP (the smaller ru), in par-ticular at P7. However, it should be noted that the ac-tual vertical stresses of PRF beneath the tank largelydepend on the raft contact pressure. The pore waterpressure behaviors of Cases 3b and 4 were almost thesame from the beginning to the end, except those ofP7 and P4 at the tank beneath showing slightly largerEPWPs in Case 4 than in Case 3b in the rapid EPWPrise period, which could be attributed to the increaseof raft pressure in this period. While for Cases 5 and 6(driven-PRFs), the EPWP behaviors in the rapid riseperiod are also similar to those in Cases 3b and 4 inalmost all locations. However, P7 shows quite a dif-ferent behavior in Cases 5 and 6 compared to thosein Case 3b and Case 4. P7 in Case 6 shows twostage EPWP build-up until t = 2 sec and t = 4.6 secand a gradual decrease immediately after the rapidbuild-up. In Case 5, P7 increased in one stage untilt = 3 sec and showed a gradual decrease from thispoint, which is different from the behavior observedin Cases 3b and 4, where EPWP increased graduallyafter the rapid rise. These different behaviors of P7in the different cases are considered as an effect ofthe pile driving. The pile driving process increased

the density and lateral confinement of sand betweenpiles, which could slower the EPWP increase and en-hance (elongate) the EPWP dissipation at the loca-tion of P7. This effect was more significant in Case6 than in Case 5 due to larger number of driven piles.This EPWP behavior also affected the pile resistance,resulting in the raft contact pressure, which will bediscussed in the next section. The later start of dis-sipation of EPWP at the locations outside of the raftin Case 6, compared to the other cases is an evidenceof smaller relative density in Case 6 (Table 1). Thelarger and longer EPWP at the perimeter area in Case6 could cause the larger t2 and t3 value, of which def-initions are subsequently explained. In addition, theEPWP behavior in the area outside of the tank borderis different in Cases 5 and 6 (driven PRFs) comparingto the other cases. Possible reasons for this differenceare slightly larger input motion in Case 5 and smallerrelative density in Case 6 than in the others (Figs. 10and 11 and Table 1). The residual EPWPs observed atthe shallow depth after dissipation was due to the tanksettlement, which caused settlement of PPT at the lo-cation beneath the tank and the settlement of PPT dueto the relatively large unit weight at the location out-side of the raft. In P7 of Cases 5 and 6, however, these

366

positive (residual) EPWPs were not observed; even ifit lowered below zero at the end of shaking in Case 5.No clear reason could be found for this, but from thefact that this negative pressure finally became almostzero with a very slow increase and while the residualEPWP occurred in Cases 5 and 6 during Shake 2, thelarge disturbance at the location of P7 or tension of

Fig.14 Excess pore water pressure ratio (ru) in Shake 1.

the PPT wire, both of which might be caused by thepile driving process, could be considered as possiblereasons.

As mentioned earlier, the typical variation of EP-WPs can be divided into three parts. The first partis until the end of rapid increase, which is called“buildup period” and t1 is the end of this period. Thesecond part is from t1 to the time when the pore pres-sure starts the dissipation (t2), which is named “liq-uefaction period”. The third is the dissipation stagefrom t2 until the end of this period (t3). Determin-ing the end of dissipation (t3) is not straightforwarddue to the gradual decrease of EPWP and the residualEPWP. The end of dissipation (t3) was determined atthe time when the decrease of EPWP from the maxi-mum value became 99% of that at the end of the mea-surement as shown in Fig. 13. These three times (t1,t2, and t3) are highlighted in the EPWP graphs of P7(Fig. 13) and will be used for the discussion on thelater parts. It should be noted that there is some un-certainty in the determination of t2 of P7 for Cases5 and 6. As the EPWP showed the gradual decreasefrom t1, t2 was determined by the time when a rela-tively large EPWP decrease rate was observed.

(3) Raft base contact pressuresFigure 15 shows the variations in raft base contact

pressures measured by five pressure cells during thetwo shakings. In Case 4, the cell at the tank center(EP5) could not measure the pressure. In the figure,ARP, which is the average of the pressures recordedby EP cells is also represented by dotted line. Themeasured pressures in Case 4 with the non-built-inpressure cells were more uniform than those mea-sured in Cases 3b, 5, and 6 with the built-in pressurecells, which was partly because of the effect of initial

Fig.15 Variations in raft base contact pressures during the shakings (*: EP5 is not included in ARP of Case 4).

367

surface undulation on the measured pressures. Theeffect was considered more significant for the built-incells, which created a flat raft base than the nonbuilt-in cells with a convex on the base. However, com-paring the variations in Cases 4 and 5 or 6, it can beconfirmed that the built-in cells could measure the dy-namic pressure better than the non-built-in cells. Alsothe variations in Case 3b shows that the average valueof the five pressures measured by built-in cells (ARP)during the shaking was close to the vertical load in-tensity (qv = 81 kPa), while the non-built-in cells inCase 4, almost all cells recorded the pressure morethan qv as observed in the preloading stage (Fig. 6).From these facts, it can be reconfirmed that the built-in cells could give more precise pressures than thenon-built-in ones for both static and dynamic load-ings. Between Cases 5 and 6, there are significantdifferences in ARP, especially in Shake 1. BeforeShake 1 the ARPs were 44 kPa and 11 kPa, respec-tively. Although there should be some difference be-tween precise average raft pressure and the measuredARP, assuming they are quite similar, the vertical loadwas carried about 54% by the raft at the beginning ofShake 1 in Case 5, while in Case 6 only 14% of thevertical load was carried by the raft. This differencein the Raft Load Proportion (RLP: ARP/qv) inevitablyoccurred due to the larger resistance of piles in Case 6with 24 piles than that in Case 5 with 12 piles. How-ever, the RLPs of the two cases became similar af-ter the first shake, 20 kPa (RLP=25%) in Case 5 and17 kPa (RLP=21%) in Case 6, respectively. In Shake2 some changes in ARP took place, but the changeswere small compared to those in Shake 1, from 20 kPato 25 kPa (RLP=31%) in Case 5 and from 17 kPa to12 kPa (RLP=15%) in Case 6.

The measured raft pressures showed even more sig-nificant changes during the shaking with different pat-terns depending on the types of foundations, the loca-tions and the shaking history, i.e., Shake 1 or Shake2. As a common trend of PRF, the pressures all in-creased in the EPWP build-up stage (till t1 of P7shown in Fig. 15). After that, the pressures showeddifferent behaviors in different cases or shakings. Forexample, in Shake 1 of Case 5, all pressures reachedto the vertical load intensity (qv) till t2 in the liquefac-tion stage, and then all pressures near the perimeterof the raft and the ARP started decreasing, but EP5at the center of the raft increased after t2. This par-ticular behavior implies that due to the liquefaction,the piles lost almost all the vertical resistance andthe vertical load was carried by the raft during theliquefaction stage. Then at the consolidation stage,the vertical load was redistributed to the inner part ofthe raft below which the soil had a relatively higherstiffness than the outer part of the raft, and the pilesregained the vertical resistance due to dissipation of

EPWPs, meaning more pile resistance mobilized thanthe initial value before the shaking especially near theperimeter.

This trend in the variations in raft base pressurecan be seen also in Shake 1 of Case 6 to some ex-tent. However, the detail behavior is quite differentespecially in the beginning. The base pressures didnot increase until about 5 sec and showed a rapid in-crease till t1 but up to about half of qv. In Case 6 withmore driven piles, the piles vertical resistance couldbe maintained in the beginning and then decreasedby the liquefaction, but about a half, not all as ob-served in Case 5. In the consolidation stage the ARPdecreased; in other words, the piles resistance wasregained, but ARP at the end of shaking was largerthan the initial value, meaning the mobilized pile re-sistance at the beginning, was not fully regained afterthe shake in Shake 1 of Case 6. It is also interest-ing that the amplitudes of dynamic component of thepressure were much larger for Case 5 than for Case 6,which could be attributed to the smaller rocking mo-tion in the latter case and the larger vertical resistanceof piles against dynamic load from the tank for Case6 than for Case 5. The increase in the base pressure(EP5) or the stress concentration at the raft center bythe shaking can be confirmed also in the slab foun-dation (Case 3b). The level of stress concentrationof SF was much larger than PRF and this concentra-tion could be decreased by increasing the pile den-sity, which was partly because of relatively less con-fined stress with less RLP and also larger total pilesresistance. Almost the same behavior as observed inShake 1 occurred in Shake 2 with a slight difference.The large pressure at EP5 showed a rapid decrease atthe beginning of Shake 2 even in SF (Case 3b). On theother hand, the increases in ARP by the liquefactionwere smaller than those of Shake 1, which was theeffect of densification by the first shaking. Further-more, in Case 6, EP5 showed a reduction after t2, thatsuggested relative large regaining of pile resistance inthe consolidation part. From these observations it isconsidered that these unstable or non-uniform pres-sures at the foundation base could be compensatedby the additional resistance from the piles. Due tothe failure in measuring EP5, the stress concentrationat the raft center could not be confirmed in Case 4.However, the overall trend in the pressure behavior atthe perimeter part of the raft are similar to the others.Also because of the uncertainty by the non-built-inpressure cells used in Case 4, the effect of pile instal-lation process on the raft base pressures could not beobserved between Case 4 and Case 5.

The ARP in Shake 1 and Shake 2 for all models areshown together with EPWP at P7 in Fig. 16. In thecase of slab foundation without piles (Case 3b), somevariation and difference from the vertical load inten-

368

Fig.16 Variation of average raft pressure (ARP) during theshakings (*: EP5 is not included in ARP of Case4).

sity (qv) of ARP were observed during the shaking.This could be a limitation of simple averaging pro-cess of measured pressures. However, the variationin ARP is not so large and is close to qv comparedto those in PRF cases. On the other hand, in PRFcases, a common trend could be confirmed from thefigure; that is, the raft load proportion increased bythe reduction of pile loads due to the liquefaction, butby the recovery of effective stresses of the soil dueto the dissipation of EPWPs the raft load decreased.In other words, the pile load was regained gradually.This raft load (ARP) reduction is slower in Case 6 incomparison to Cases 4 and 5, because the EPWP dis-sipation stage (t3 − t2) was longer in Case 6 as shownin Fig. 16. The recovery of pile load was earlier inShake 2 than in Shake 1, which corresponds to thefact that t2 in Shake 2 was earlier than that in Shake1 for these cases (Fig. 16). Comparing the behav-ior in the EPWP build-up stage until t1, only Case 6showed a clearly different behavior between Shake 1and Shake 2. In Shake 1 of Case 6, EPWP beneaththe tank (P7) showed relatively slow and two-stagesincrease as discussed before (Fig. 13), and the ARPdid not increase until about 4.5 sec, which could bean effect of large number of pile driving. However,in Shake 2, the quick rise occurred in 3 sec both in

EPWP and ARP. From the behavior in Shakes 1 and2, it can be concluded that the effects of pile drivingby the static penetration, such as the increase in hori-zontal stress, could be eliminated by the shaking.

The behavior of piled raft foundation also dependson the frictional resistance of raft base. As explainedbefore, silica sand No. 8 was glued on the bottom sur-face of the raft model to create a rough surface con-dition. The frictional resistance of tank base on thesand surface is τ f = σ× tg(ϕ′) = 68 kPa and the max-imum of dynamic horizontal stress on the raft surfaceis τ = (m × amax)/A = 24 kPa. In these equations,σ is tank normal stress on the subsurface (qv = 81kPa), ϕ′ is angle of friction (40◦), m is mass of tank(363 ton), amax is maximum horizontal accelerationof tank (0.3 g), and A is raft base area. As this simplecalculation indicates, the dynamic horizontal stress ismuch smaller than tank frictional resistance. There-fore, any slip between the tank and subsoil is not tobe expected and the performance of PRF might notbe affected.

(4) Tank responsea) Tank accelerations and rocking motion

The accelerations at the top and bottom of the tankin the shaking direction (A9, A8) are shown withthe input acceleration in Fig. 17. In the figure, thetank accelerations in the transverse direction at the top(A13) and the bottom (A12) are also shown, but A13could not be recorded in Case 6. The Fourier ampli-tude spectra of the tank accelerations both in the shak-ing and transverse directions are presented in Fig. 18.In the shaking direction during the rapid increase inEPWP in the build-up period before t1, the differencebetween the tank bottom and top accelerations wererelatively small. However, after t1, the bottom andtop accelerations showed a difference; bigger at thetop and smaller at the bottom than the input, whichcan be considered as an evidence of the rocking mo-tion of the tank. Case 6 had the least difference be-tween the top and bottom accelerations. In particular,the spectra of the tank top and bottom accelerationswere similar to that of input. This is a clear evidenceof the effectiveness of higher pile density in reducingthe rocking motion of the tank. In Cases 3b, 4, and5, short period components were significantly atten-uated at the tank bottom, but the long period com-ponents amplified. As a difference between SF andPRFs, it can be observed that the phase differencesbetween the tank bottom and top accelerations weremore significant in the late part of shaking in Case3b (SF) than in Cases 4, 5, and 6 (PRFs). The tankrocking motion was estimated using the recorded ver-tical displacements by L1 and L2 at two sides of thetank. For this purpose, first, the recorded displace-ments were smoothed to have a static component of

369

Fig.17 Tank response accelerations in Shake 1.

Fig.18 Tank response Fourier spectrum in Shake 1.

370

Fig.19 Tank dynamic rotation in the shaking direction observed by vertical displacements of tank during Shake 1.

Fig.20 Tank center settlements, top: entire shaking with t2, bottom: earlier part of shaking with t1.

the displacement and the smoothed data were sub-tracted from the measured displacements, the remain-ing can be considered as dynamic part of displace-ments at the tank edges (L1d and L2d). The differencebetween L1d and L2d (L1d − L2d) divided by the dis-tance of the two LDTs is considered as the dynamiccomponent of the tank rotation in the shaking direc-tion, which is shown in Fig. 19. The amplitude of therotation was the smallest for Case 6 and both Cases4 and 5 had almost the same amplitudes, which werecompatible with the results presented in Fig. 18. Thelargest maximum amplitude was observed in Case 3bamong all cases. Although no significant differencein the tank accelerations can be seen between Case 3band Cases 4 and 5 in Figs. 17 and 18, the dynamicrotation shown in Fig. 19 indicates the effectivenessof PRFs in reducing the tank rocking motion even ina small number of piles.

In the transverse direction, the accelerations weremuch smaller than those in the shaking direction es-pecially at the tank bottom with negligible amplitudeand no clear dominant period. At the tank top in Cases3b and 4, some vibration with maximum value about20% of the acceleration in the shaking direction wasobserved. The dominant period of tank top vibrationin the transverse direction was about 0.3 sec, which

was half of that in the shaking direction (0.6 sec). Thisspecific response of the tank top in the transverse di-rection could be attributed to the deflection of tanktop due to the relatively small hoop stiffness at the toppart of the tank wall, while the tank bottom was muchstiffer than the top due to the aluminum-made raft,which was considered a rigid plate. Similar deforma-tions might be developed due to the inertial force anddynamic water pressure inside the tank at both timeswhen these forces acted in the positive and negativedirections. The flexible deformation of the tank couldalso cause the acceleration difference at the tank bot-tom and top in the shaking direction. This accelera-tion difference could be one of the reasons for incon-sistency in the discussion of rocking motion done bythe tank accelerations and the dynamic rotation.b) Tank settlements

In Fig. 20, the settlements at the tank center, whichis the average of L1 and L2 are compared for the entireperiod and the early stage of shaking at the top andbottom figures. The t1 and t2 obtained from P7 andP8 are also indicated in the figure. In Case 6, becauseP8 could not be recorded, the times of P9 are substi-tuted. The settlements increased gradually during theshaking in contrast to the behavior of EPWPs, that is,a quick rise within a short time (Fig. 13). Compar-

371

ing the results in Shake 1 and Shake 2, the effect ofdensification by the first shake could be confirmed.Even though the input motion in Shake 2 was largerthan that of Shake 1, the settlements in the secondshake were smaller than those of the first. The ef-fect of pre-shaking was more significantly evidencedin the beginning of the shaking. In Shake 1 of Cases3b, 4, and 5, the tank started settling at the time ofabout 1.5 sec, which was the actual onset of the shak-ing in terms of Arias intensity (Fig. 11), and the set-tlement rate increased with time until t = 3 sec whilein Shake 2, there were no substantial settlement untilt = 2.3 sec. The relatively large settlement of Shake 1in the beginning of shaking could be attributed to thepoor contact of the raft base to the ground surface, akind of bedding error, which can be removed by thefirst shake. After this initial part, the settlement in-creased almost linearly with time until t = 8 sec atthe time when EPWP dissipation started in the deepdepth beneath the tank (P2 as shown in Fig. 13). InShake 1 of Case 6, the settlement started with delay at2.5 sec and increased slowly until about 4.5 sec but af-ter that, it increased rapidly until 8 sec, the time whenthe dissipation in the deeper part started. This delayand slow increase in the settlement at the beginningcan be attributed to the slow EPWP increase and theremaining pile resistance as the effects of high densityof driven piles as discussed before (Fig. 16). How-ever, after 4.5 sec the effect could be diminished bythe shaking, which resulted in further EPWP genera-tion and the reduction of piles resistance and the largesettlement. Although the settlement rate started de-creasing at 8 sec, further settlement occurred even af-ter the time when EPWP just beneath the raft starteddecreasing (t2P7), until the time when EPWP at theshallow depth beside the raft started decreasing (t2P8

or t2P9 (Case 6)). After this time, the minor settlementtook place, which was mainly caused by small shak-ing and the consolidation of sand. The residual set-tlement in the late stage of the tests seems smaller forCases 4, 5, and 6 (PRFs) than for Case 3b (SF). Re-covery of pile bearing load, which can be confirmedin Fig. 16, could be a reason for the smaller settlementin the late stage of the shaking and after the shaking inthe PRFs than in the SF. In Shake 2, the particular dif-ference between Case 6 and the other cases observedin Shake 1 was no longer observed, which was alsoan evidence that the effects of pile driving could bediminished by the first shaking.

As for the overall effect in reducing the settlement,simple direct comparisons could not be done becauseof different input acceleration and initial density ofthe sand. To discuss these effects, the EPWP behav-ior could be a reference, especially liquefaction time(t2) and consolidation time (t3) at the ground outsideof the tank. This is because the liquefaction time and

the consolidation time after the liquefaction will belonger for the less dense sand and the larger shakingif the shaking time is the same. From Fig. 13, the levelof liquefaction is considered the highest in Case 6 andsecond in Case 5, while Cases 3b and 4 had almost thesame liquefaction level. Considering the relative sig-nificance of the liquefaction or shaking, the effect ofpiles of PRF as settlement reducers can be confirmedfrom the results of Case 3b and Case 4. While for theeffects of pile driving and the pile number, they canbe seen in the beginning of Shake 1 between Cases 4and 5 and Cases 5 and 6 respectively. However, theeffects of pile driving and pile number on the settle-ment reduction cannot be confirmed in the later partof Shake 1 and in Shake 2 due to the elimination ofthe driving effects by the shaking and incomparable-ness of the test conditions.

Besides the above-mentioned effects, i.e., sand den-sity, input motion, and pile installation method, loadproportion between the raft and pile could be a criti-cal factor. Before Shake 1 of Case 6, the majority ofthe vertical load was carried by the piles (Fig. 16) andthe settlement was very small before losing the pileresistance till t ∼4.5 sec. But once the piles lost resis-tance and the load was transferred to the raft, a largesettlement occurred even for the case with 24 piles.

c) Tank maximum rotationIn the safety assessment of tank foundation, the un-

even settlement is a critical concern. For the relativelysmall diameter tank supported by a rigid slab or raft,the uneven settlement is equivalent to the rotation ofthe foundation. In the previous dynamic model testsof foundation using the one-directional shaking table,e.g., Takemura et al. (2014),20) the rotation of tankfoundation was only measured in the shaking direc-tion. In this study, with the settlement at three lo-cations (Figs. 1 and 22), the maximum rotation wasmeasured.

Figure 21 shows the variation in the maximum ro-tation during the shaking for the entire period andearly stage of the shaking in the top and bottom fig-ures with t1 and t2 of P7 and P8 or P9 (Case 6), re-spectively. The maximum rotation is calculated usingequation of flat plane in geometry and the recordeddata of three LDTs at the top of the tank (L1, L2, andL3). Assuming the location of three LDTs as hori-zontal coordinate (x and y) and the settlement at threelocations as the vertical coordinate, the point of themaximum settlement on the raft (point D shown inFig. 22) was estimated. Then the maximum rotationof tank in the direction from tank center to point Dwas calculated. In this calculation, the three LDTsdata were used after eliminating the high-frequencycontents in order to remove the noise from the cal-culated results especially in the beginning when thesettlements were relatively small. In the beginning

372

Fig.21 Tank maximum rotation, (a) entire shaking with t2(b) earlier stage of shaking with t1.

Fig.22 Variation in direction of tank maximum rotation.

of shaking, the rotation gradually increased with timebut was very small until t1 of P7 in all the cases ex-cept that of Case 6 (Driven PRF, 24 Piles). Althought1P7 in Shake 1 of Case 6 was longer than those ofthe other cases, the increase in rotation in the earlystage was much larger both for Shake 1 and Shake 2.In Shake 1, the rapid increase in the rotation startedfrom t = 3.5 sec, but in Shake 2 the increase startedfrom t = 2 sec. The time of the large rapid rotationincrease in Shake 1 of Case 6 was close to the time ofthe onset of the ARP increase from a very small value(Fig. 16) and the time of the onset of increase in the

settlement (Fig. 20). It is considered that in the be-ginning of Shake 1 in Case 6, the almost vertical loadwas carried by piles and the raft base contact condi-tion to the ground surface could not be uniform withsome local gaps due to the small raft load. This initialcondition caused thelarge and relatively non-uniformsettlement by the reduction of pile resistance, result-ing in very large rotation from the beginning. Theearlier large rotation in Shake 2 of Case 6 could be aresult of the initially existing inclination of the foun-dation. This large settlement and rotation could alsobe enhanced by the relatively large level of liquefac-tion in Case 6 as compared to the others as discussedin the previous section.

Besides Case 6, the behaviors of the tank on lique-fied sand in the other cases were also quite compli-cated and different in different cases. In Case 4 (non-driven PRF, 12 Piles), the rotation increased mono-tonically in the liquefaction stage (from t1 to t2) bothfor Shake 1 and Shake 2. While in Shake 1 of Case 3b,the rotation increased after t1 but tended to decreasein the liquefaction period. In Shake 2 of Case 3b, therotation behavior fluctuated more during the shaking.In both shakes of Case 5, the rotation increased mono-tonically in the liquefaction period but not as much asCases 4 and 6. As a result, the tank rotations after theshaking were larger for Cases 4 and 6 (PRFs) than forCase 3b (SF) and Case 5 (PRF). The large dynamicraft base pressure amplitudes can be pointed out as acommon behavior of Case 3b and Case 5 as comparedto relatively small amplitudes in Case 6 (Fig. 15) es-pecially in Shake 1. This large reaction from the raftagainst the dynamic load could also positively workfor reducing the rotation.

Figure 22 shows variations in the direction of max-imum rotation, θ, during the shaking. The definitionof θ is given in Fig. 22. From the figure, the PRFcases could be divided into two groups. In Cases 4and 6, the direction suddenly changed from the shak-ing direction (θ = 0 or 180◦) to a different directionand became constant to a certain direction more trans-verse than the shaking direction. While in Case 5,although some fluctuation in the direction took placein the beginning, the maximum rotation directed tothe shaking direction (θ ∼ 180◦). On the other hand,in Case 3b, the direction showed unstable behavior,gradually changing from the shaking direction untilthe end of the shaking, which corresponded to thechange in rotation during the shaking shown in Fig.21. Comparing Figs. 21 and 22, it can be seen alsothat the rotation tends to be larger for the case withmaximum rotation direction more to the transverse di-rection than to the shaking one. From these observa-tions, it can be inferred that once the direction of therotation is fixed to a direction diverted from the shak-ing direction, the rotation will be accumulated by the

373

Fig.23 Tank maximum rotation vs settlements.

Fig.24 Peak of tank acceleration vs peak of water pressureincrement.

shaking. But while the rotation mainly takes placein the shaking direction in the beginning of shaking,the monotonic increase in the rotation may not easilyoccur and the rotation behavior becomes very compli-cated as seen in the slab foundation of Case 3b (Fig.21). This unstable behavior could be prevented bythe additional support from the piles of PRF in Case5. There should be several reasons why Cases 4 and6 (PRFs) tilted in the diverse direction, such as in-evitable difference in bearing resistance of each pile,and non-uniformity of raft base contact condition tothe ground, which could cause the rotation of the tanktoward the area with small pile resistances and poorcontact of the raft base to the ground.

The relationship between the tank maximum ro-tation and the tank center settlement is presented in

Fig. 23. The relationships are shown along with themarks at the time of EPWP buildup (t1) and liquefac-tion stage (t2) obtained from the location of P8 or P9(Case 6) at the shallow depth beside the tank. Thesetimes are considered as indicators of the partial lique-faction and complete liquefaction. The relationshipsare very different for the various cases and betweenthe first and second shakings. As an overall trend ofthe relations, the following can be pointed out: (1)majority of the settlement and rotation took place inthe liquefaction stage except in Case 6; (2) in the earlystage until t1, the rotations of PRFs except for Case 6,were smaller than or equal to those of SF; (3) neglect-ing the large settlement and tilting caused by poor raftcontact in Case 6 (due to the small ARP at the begin-ning of shakings (Fig. 16) and some unevenness inthe raft contact with subsoil (EP4 in Fig. 9), the effec-tiveness of driven PRF compared to non-driven PRFsin reducing uneven settlement can be seen from theresults of Cases 5 and 4.d) Sloshing of liquid inside the tank

A PPT was located inside the tank at a corner (Fig.1), to record the water level change during the shak-ing. The peak of water pressure increments (waterpressure -initial water pressure before shaking) andthe peak of acceleration of tank at center of gravity ineach 2 seconds interval of shaking are measured anddrawn against each other in Fig. 24. As the figureshows, the water pressure increment is related to thetank acceleration. In the other words, water heightis controlled by the acceleration response of tank re-gardless of the foundation type.

4. CONCLUSIONS

In this study, a specific modeling technique was de-veloped to install driven piles into a model groundstatically under in-flight conditions of centrifuge tomodel piled raft foundations with different numbersof driven piles (Driven PRFs). A slab foundation (SF)

374

and a PRF with non-driven piles (Non-driven PRFs)were also made. The following conclusions were de-rived from the dynamic centrifuge model tests on atank model supported by these foundations on lique-fiable saturated sand.

(1) The pile driving process could slow down the ex-cess pore water pressure (EPWP) increase andenhance the EPWP dissipation beneath the tankespecially in the case with large number of piles.The slow EPWP increase and remaining pile re-sistance as the effects of high density of drivenpiles caused delay and slow increase of the tank’ssettlement in the beginning. However, the effectof pile driving was eliminated by the liquefac-tion in the entire depth of piles because the lateralstress increased by the static pile driving was di-minished during the liquefaction.

(2) The raft base contact pressures significantlychange during the shaking. As a common trend,the base contact pressures at all location increasein the EPWP build-up stage. At the outer part ofthe raft base, the pressures are kept almost con-stant during the liquefaction stage and then startdecreasing in the consolidation stage; while inthe central part, the pressure increases even inthe consolidation stage, showing the stress con-centration due to the relatively large stiffness bythe raft pressure at the central portion. The in-crease in raft load proportion (RLP) of PRF iscaused by the reduction of piles resistance dueto the liquefaction, but the RLP decreases grad-ually with the recovery of piles resistance by theEPWP dissipation.

(3) The reduction of pile resistance could be slowedby the pile driving to some extent, but the effectcould not be expected after a large shaking. Thechange in RLP during and after the shaking de-pends on the number of piles and level of lique-faction. The lesser the pile number and the morethe liquefaction significance are, the larger theRLP during the liquefaction and the slower theRLP reduction after the liquefaction.

(4) The piled raft foundation could reduce the rock-ing motion of tank during the shaking in com-parison to the slab foundation, especially for theDriven PRF with large number of piles.

(5) The piled raft foundation is effective in reduc-ing tank settlement compared to the slab founda-tion. But relatively large raft load proportion anduniform contact condition of raft and subsoil arecritical conditions for preventing a large unevensettlement of the foundation.

(6) The dynamic behavior of pile raft foundation onliquefiable sand is so complicated, which is af-fected by various factors. For the conditionsof complete liquefaction in the entire ground,

a better performance of PRF could not be ex-pected. Therefore the effectiveness and limita-tion of PRFs should be studied more for differ-ent ground conditions, i.e., the ground with par-tial liquefaction, such as, the liquefaction in par-tial depth or the ground with non-liquefiable soillayer, and different pile conditions, such as, thedynamic or static pile installation methods andthe pile head fixity, which are all critical for therational application of PRF to oil storage tanks.

REFERENCES1) Ishihara, K., Kawase, Y. and Nakajima, M. : Lique-

faction characteristics of sand deposits at an oil tanksite during the 1978 Miyagiken-Oki Earthquake, Soilsand Foundations, Vol. 20, No. 2, pp. 97-111, 1980.

2) Cubrinovski, M. and Ishihara, K. : Analysis of theperformance of an oil-tank pile foundation in liq-uefied deposits, Proc. of XV ICSMGE TC4 SatelliteConference on “Lessons Learned from Recent StrongEarthquakes”, 2001.

3) Burland, J. B., Broms, B. B. and DeMello, V. F. B. :Behaviour of foundations and structures, Proc. of 9thICSMFE, 1977.

4) Horikoshi, K. and Randolph, M. F. : A contribution tooptimum design of piled rafts, Geotechnique, Vol. 48,No. 3, pp. 301-317, 1998.

5) Poulos, H. G. : Piled raft foundations: design and ap-plications, Geotechnique, Vol. 51, No. 2, pp. 95-113,2001.

6) Yamashita, K., Yamada, T. and Hamada, J. : Inves-tigation of settlement and load sharing on piled raftsby monitoring full-scale structures, Soils and Founda-tions, Vol. 51, No. 3, pp. 513-532, 2011.

7) Yamashita, K., Hashiba, T., Ito, H. and Tanikawa,T. : Performance of piled raft foundation subjectedto strong seismic motion, Geotechnical EngineeringJournal of the SEAGS & AGSSEA, Vol. 45, No. 2,2014.

8) Pastsakorn, K., Hashizume, Y. and Matsumoto, T. :Lateral load tests on model pile groups and piled raftfoundations in sand, Proc. of International Confer-ence on Physical Modelling in Geotechnics, 2002.

9) Matsumoto, T., Fukumura, K., Horikoshi, K. andOki, A. : Shaking table tests on model piled rafts insand considering influence of superstructures, Inter-national Journal of Physical Modelling in Geotech-nics, Vol. 4, No. 3, pp. 21-38, 2004.

10) Horikoshi, K., Matsumoto, T., Hashizume, T., Watan-abe, T. and Fukuyama, H. : Performance of piled raftfoundations subjected to static horizontal loads, Inter-national Journal of Physical Modelling in Geotech-nics, Vol. 3, No. 2, pp. 37-50, 2003.

11) Sawada, K. and Takemura, J. : Centrifuge model testson piled raft foundation in sand subjected to lateraland moment loads, Soils and Foundations, Vol. 54,No. 2, pp. 126-140, 2014.

12) Horikoshi, K., Matsumoto, T., Hashizume, Y. andWatanabe, T. : Performance of piled raft foundationssubjected to dynamic loading, International Journalof Physical Modelling in Geotechnics, Vol. 3, No. 2,pp. 51-62, 2003.

375

13) Notification of Technical Specifications Regardingthe Regulations on Hazardous Materials : FDMA(Fire Disaster Management Agency, Japan), 1974.

14) Fellenius, B. H. and Ochoa, M. : Large liquid stor-age tanks on piled foundation, Proceedings of Inter.Conf. on Foundation and Soft Ground Engineering—Challenges in the Mekong Delta, pp. 3-17, 2013.

15) Sento, N., Yasuda, S., Yoshida, N. and Harada, K. :Case studies for oil tank on liquefiable sandy groundsubjected to extremely large earthquakes and counter-measure effects by compaction, Proceedings of 13thWorld Conference on Earthquake Engineering, 2004.

16) Liew, S., Gue, S. and Tan, Y. : Design and instrumen-tation results of a reinforcement concrete piled raftsupporting 2500 ton oil storage tank on very soft allu-vium deposits, Proc. of the 9th Inter. Conf. on Pilingand Deep Foundations, 2002.

17) Chaudhary, M. T. A. : FEM modelling of a large piledraft for settlement control in weak rock, EngineeringStructures, Vol. 29, No. 11, pp. 2901-2907, 2007.

18) De Sanctis, L. and Russo, G. : Analysis and perfor-mance of piled rafts designed using innovative crite-ria, Journal of Geotechnical and GeoenvironmentalEngineering, Vol. 134, No. 8, pp. 1118-1128, 2008.

19) Imamura, S., Yagi, T. and Takemura, J. : Dynamic sta-bility of oil tank supported by piled-raft foundation onliquefiable sand, Proc. of the 7th International Con-ference on Physical Modelling in Geotechnics, Vol. 2,pp.1409-1414, 2010.

20) Takemura, J., Yamada, M. and Seki, S. : Dynamicresponse and settlement behavior of piled raft foun-dation of oil storage tank, Proc. of International Con-ference on Physical Modelling in Geotechnics, Vol. 1,pp. 613-619, 2014.

21) Adejumo, T. W. and Boiko, I. L. : Effect of installa-tion techniques on the allowable bearing capacity ofmodeled circular piles in layered soil, InternationalJournal of Advanced Technology & Engineering Re-

search, Vol. 2, No. 8, pp. 1536-1542, 2013.22) Flynn, K. N. and McCabe, B. A. : Shaft resistance of

driven cast-in-situ piles in sand, Canadian Geotechni-cal Journal, Vol. 53, No. 1, pp. 49-59, 2015.

23) Bloomquist, D., Feld, T., Townsend, F. C., Grav-gaard, J. and Gill, J. : Development of a multiple piledriver/load test device for pile group studies, Proc. ofthe Centrifuge 91, pp. 355-359, 1991.

24) Pan, S., Pu, J., Yin, K. and Liu, F. : Developmentof pile driver and load set for pile group in centrifuge,Geotechnical Testing Journal, Vol. 22, No. 4, pp. 317-323, 1999.

25) De Nicola, A. and Randolph, M. F. : Centrifuge mod-elling of pipe piles in sand under axial loads, Geotech-nique, Vol. 49, No. 3, pp. 295-318, 1999.

26) Sahraeian, S. M. S., Takemura, J. and Sakae, S. :A study on seismic behavior of piled raft founda-tion for oil storage tanks using centrifuge model tests,Proc. of 6th International Conference on EarthquakeGeotechnical Engineering, 2015.

27) Takemura, J., Yamada, M. and Seki, S. : Dynamic re-sponse and settlement behavior of piled raft founda-tion of oil storage tank, Proc. of Joint 9th Inter. Conf.of CUEE/4th Asia Conf. on Earthquake Eng., pp.533-544, 2012.

28) Das, B. M. : Fundamentals of Geotechnical Engineer-ing, 3rd edition, Chris Carson, 2007.

29) Ishimatsu, S., Yagi, T., Yoshimi, T. and Takemura, J.: Filed observation of pile behavior during the liquidlevel variation in an oil tank, Kisoko, Vol. 37, No. 10,pp. 76-79, 2009. (in Japanese)

30) http://www.jma.go.jp/jma/indexe.html, JMA (JapanMeteorological Agency), 2008.

31) Kramer, S. L. : Geotechnical Earthquake Engineer-ing, Pearson Education, 1996.

(Received November 14, 2016)

376