otechnique 63, No. 14, 12301244 [http://dx.doi.org/10.1680/geot.12.P.149] Al-Defae, A. H. et al. (2013).

Ge

Aftershocks and the whole-life seismic performance of granular slopes

A . H . A L - D E FA E , K . C AU C I S a n d J. A . K NA P P E T T

Shallow embankment slopes are commonly used to support elements of transport infrastructure in seismic regions. In this paper, the seismic performance of such slopes in non-liqueable granular soils is considered, focusing on permanent movement and dynamic motion at the crest, which would form key inputs into the aseismic design of supported infrastructure. In contrast to previous studies, the evolution of this behaviour under multiple sequential strong ground motions is studied through dynamic centrifuge, numerical (nite-element, FE) and analytical (sliding-block) modelling, the centrifuge tests being used to validate the two non-physical approaches. The FE models focus on the specication of model parameters for existing non-linear constitutive models using routine site investigation data, allowing them to be used routinely in design and analysis. Soil-specic constitutive parameters are derived from shearbox and oedometer test data, and are found to signicantly outperform existing empirical correlations based on relative density, highlighting the importance of specifying a suitably detailed site investigation. An improved sliding-block (Newmark) approach is also developed for estimating permanent deformations during preliminary design, in which the formulation of the yield acceleration is fully strain-dependent, incorporating both material hardening/ softening and geometric hardening (re-grading). The site-specic (improved) FE models and the new sliding-block approach are shown to outperform considerably existing FE parameters and sliding-block models in capturing the permanent deformations of the slope under virgin conditions, and further, only the improved FE and sliding-block models are found to capture correctly the behaviour of the damaged slope under subsequent earthquakes (e.g. strong aftershocks). The FE models can additionally accurately replicate the settlement prole at the crest and quantify the dynamic motions that would be input to supported structures, although these were generally overpredicted. The FE procedures and sliding-block models are therefore complementary, the latter being useful for preliminary design and the former for later detailed design and analysis.
KEYWORDS: centrifuge modelling; earthquakes; embankments; numerical modelling; sands; slopes

INTRODUCTION Transport infrastructure is a vital lifeline that must be safeguarded in seismic regions. Shallow embankment slopes and cuttings will commonly be used along the length of a road or railway line to allow changes in gradient, and damage to this type of infrastructure could inhibit the movement of emergency services and rebuilding in the aftermath of an earthquake. Although the infrastructure generally supported on such constructions is relatively light (low bearing pressure) and exible, signicant damage can be caused by large permanent seismic slip within the slope. A popular method of analysis for predicting seismic slip is the Newmark sliding-block technique (Newmark, 1965). In this method a yield acceleration is calculated under pseudostatic conditions using limit analysis techniques for simple cases (e.g. Kim & Sitar, 2004), or numerical methods such as nite-element limit analysis (FELA) for more complex cases (e.g. Loukidis et al., 2003). A strong ground motion (either a historically recorded motion or a synthetically produced accelerogram) is then used to determine the net downslope acceleration, which can then be integrated numerically to obtain slip velocity and slip displacement (Jibson, 1993). Since it was originally proposed in the 1960s, this method has been substantially improved to incorporate
Manuscript received 5 October 2012; revised manuscript accepted 7 June 2013. Published online ahead of print 15 July 2013. Discussion on this paper closes on 1 April 2014, for further details see p. ii. University of Dundee, UK, and Wasit University, Iraq. y University of Dundee, Dundee, UK.

material strength characteristics, which may be either strainhardening or strain-softening (Matasovic et al., 1997; Wartman et al., 2005), and to incorporate the dynamic behaviour (e.g. ground motion amplication) within the sliding mass itself in the case of deep rotational slips, as may be common in landll slopes (Rathje & Bray, 1999; Wartman et al., 2003). Experimental work to validate such methods has been conducted using 1g shaking tables on cohesive soils (Wartman et al., 2005); however, there remains a need to demonstrate the approachs validity at realistic stress levels (e.g. by conducting centrifuge modelling). Although such analytical tools are useful for parametric/comparative study, particularly in preliminary design, they are not able to predict the dynamic ground motions or settlement distribution (angular distortion) behind the slope crest, both of which may have a controlling inuence on the design of any infrastructure supported at the top of the slope. They are therefore not well suited to detailed analysis or design. Recent periods of seismic activity in Japan and New Zealand (20112012) have demonstrated that civil engineering infrastructure may be subjected to several successive strong ground motions within a short period (short here meaning that there has been insufcient time to complete remediation or reinstatement of damage caused by earlier ground shaking). This means that there is a further need to understand the behaviour of seismically damaged infrastructure under successive periods of ground shaking. This feature is not currently incorporated in existing sliding-block methods. This paper will consider the behaviour of shallow cohesionless slopes under a sequence of strong earthquake

1230

AFTERSHOCKS AND THE WHOLE-LIFE SEISMIC PERFORMANCE OF GRANULAR SLOPES ground motions, and develop improved analysis tools for the prediction of such behaviour. These tools will be applicable to the analysis of seismically damaged slopes under strong aftershocks, and to determination of the whole-life seismic performance of slopes (i.e. under a range of successive motions of different strengths that the slope may see during its design life). Fully dynamic numerical modelling in the time domain, using the nite-element method (FEM), will rst be applied to the problem, with the aim of producing a single analysis in the time domain that captures both the dynamic vibration effects and the permanent slope deformations. Existing nonlinear soil constitutive models will be used that encapsulate the strain history (seismic memory) of the soil. Emphasis will be placed on the efcient parameterisation of such a model: (a) using previously published correlations based on databases of element test data (which require only relative density as an input); and (b) through the use of routine laboratory test data to produce improved soil-specic calibrations. The ability to model soil response realistically, without requiring an excessive number of empirical parameters derived from non-standard tests, will allow the validated FEM procedures developed to be used with condence in routine geotechnical practice. An improved sliding-block method is also developed that can fully capture the strain history of the soil, particularly by incorporating the effects of slope deformation (re-grading) during slip within the formulation of the yield acceleration. This allows the slip of a seismically damaged (deformed) slope to be predicted analytically under subsequent strong ground motion. Both the FEM and sliding-block models are validated against centrifuge test data. The computationally simpler sliding-block model, which requires no specialist software, can then be used to undertake rapid parametric analyses

1231

(such as may be necessary during the early stages of design, or in the immediate aftermath of an earthquake) or used within a probabilistic, performance-based earthquake engineering framework if desired (Kramer, 2008). The FEM approach can subsequently be used to undertake a more detailed analysis of specic cases, which can incorporate the dynamic behaviour of any infrastructure located at the slope crest. CENTRIFUGE MODELLING Dynamic centrifuge testing was conducted using the 3.5 m diameter beam centrifuge and servo-hydraulic earthquake simulator at the University of Dundee. Two models were own, representing identical slopes of 0  288 ( 1:2) at 1:50 scale in dry sand at 50g. This slope angle was selected 9s , the to ensure that the soil was statically stable (0 , c critical state friction angle of the soil), but with a sufciently low factor of safety (and therefore low yield acceleration, khy ) to ensure that large slip displacements would be generated during strong ground motion, such that the FEM and sliding-block models could be validated to large displacements. All subsequent dimensions and properties are given at prototype scale at 50g, unless otherwise stated. The arrangement and instrumentation of the slope models is shown in Fig. 1(a). The slopes were prepared at a relative density of ID 5560% (the range accounts for the accuracy in being able to measure and replicate ID ), were 8 m tall from toe to crest, and were underlain by a further 6 m of sand at the same relative density. HST95 (Congleton) silica sand was used, which has the basic properties given in Table 1 (after Lauder, 2011). This uniformly rounded sand is very ne, and has been used by other researchers at Dundee to study other seismic phenomena (e.g Bertalot & Brennan, 2012). The values of emax and emin reported in Table 1 were

LVDTs 1 5 9 7 8 6 10 15 15 (300) Accelerometers (a) Size of model in centrifuge Fig. 1(a) 14 11 12 13 75 (150)

14 (280)

2 3 4 11 (220)

Positive acceleration (downslope)

Absorbent boundary (Lysmer & Kuhlmeyer, 1969) (b)

Fig. 1. Slope conguration: (a) centrifuge model layout and location of virtual instruments in numerical models; (b) nite-element mesh, showing boundary conditions

6 (120)

8 (160)

1232

AL-DEFAE, CAUCIS AND KNAPPETT

Magnitude: g/Hz
150 100 50 0

Table 1. State-independent physical properties of HST95 silica sand (after Lauder, 2011) Property Specic gravity, Gs D10 : mm D30 : mm D60 : mm Cu Cz Maximum void ratio, emax Minimum void ratio, emin Value 2.63 0.09 0.12 0.17 1.9 1.06 0.769 0.467

05

10

15

Magnitude: g/Hz

20 25 30 Frequency: Hz (a)

35

40

45

50

40 30 20 10 0

05

10

15

20 25 30 Frequency: Hz (b)

35

40

45

50

determined in accordance with BS1377 Part 4 (BSI, 1990). The sand was pluviated in air using a slot pluviator into an equivalent shear beam (ESB) container having exible walls, the construction of which is described in Bertalot (2012). This container was used to reduce potential boundary effects due to shear-wave reection at the container walls. The model was instrumented within the soil by 15 type ADXL78 MEMS accelerometers (70g range) manufactured by Analog Devices, and four external linear variable differential transformers (LVDTs), with a pair of instruments measuring settlement at and behind the crest of the slope along the centreline of the model, and a further pair placed adjacent to one of the side walls to measure any boundary effects (this latter pair will not be discussed further in this paper). All settlements shown subsequently in this paper are taken from the rightmost instrument in Fig. 1(a). Ground motions were applied to the models using the Actidyn QS67-2 servo-hydraulic earthquake simulator recently installed at the University of Dundee. The performance of this actuator is described in Bertalot et al. (2012). Earthquake ground motions were downloaded from the PEER (Pacic Earthquake Engineering Research) NGA database. One of the models (test AA01) was subjected to a horizontal strong ground motion recorded during the Mw 7.6 Chi-Chi earthquake in 1999 (Station TCU072, PGA 0.41g), and a second (test AA02) was subjected to a horizontal motion recorded at the Nishi-Akashi recording station in the Mw 6.9 Kobe earthquake in 1995 (PGA 0.43g). These motions had approximately the same peak acceleration, but different characteristics in the time and frequency domains, as shown in Figs 2(a) and 2(b) and Figs 3(a) and 3(b). Both records were recorded in ground with Vs . 450 m/s, representing shaking from stiffer layers beneath the soil prole tested, such that any site amplication occurs solely from the soil layer tested in the model. The former motion was used, as the Chi-Chi earthquake caused a particularly high number of slope failures (Khazai & Sitar, 2004), and is strongly directional (in these tests, the
Acceleration: g

Fig. 3. Input bedrock motions in frequency domain: (a) Chi-Chi (test AA01); (b) Kobe (test AA02)

stronger shaking was directed in the downslope direction, which is represented by positive values of acceleration). The latter motion is well known to be particularly destructive to civil engineering infrastructure having a broad frequency band below 3 Hz. The demand motions were bandpassltered between 0.8 Hz and 8 Hz (40400 Hz at model scale) using a zero-phase-shift digital lter to remove components of the signal that were outside the range that can be accurately controlled by the earthquake simulator. In both tests, four nominally identical earthquake motions were applied to the model in succession, to investigate the behaviour of the slope under strong aftershocks following initial strong shaking causing substantial slip. Details of the model tests are summarised in Table 2. All ground motions were initially calibrated on a dummy model identical to that shown in Fig. 1(a), but without instrumentation, to train the programmable logic controller within the earthquake simulator to achieve a faithful and repeatable replication of the demand motions. As a result, the ground motions applied in each successive earthquake are felt to be as close to identical as could realistically be achieved in practice, this being an idealisation of the successive motions having the same source (depth, faulting mechanism and position). FINITE-ELEMENT MODELLING The centrifuge tests were modelled using PLAXIS 2D in plane strain with the mesh and boundary conditions shown in Fig. 1(b). Compared with the centrifuge model shown in Fig. 1(a), the dimensions of the model domain were extended laterally and combined with non-reecting boundary elements controlling the dynamic stresses along the vertical boundaries (after Lysmer & Kuhlmeyer, 1969) to represent semi-innite soil conditions, that is, boundary deformations at the location of the centrifuge container wall that are controlled by the dynamic deformation of the adjacent soil. This boundary condition can also be modelled by horizontal node-to-node ties between the two vertical boundaries of a model the width of the soil tested in the centrifuge. Compared with this alternative, the method used has a higher element requirement for the same mesh density, but allows
Table 2. Summary of centrifuge models tested

05 0 05 0 5 10 15 20 25 30 Time: s (a) 35 40 45 50

Acceleration: g

05 0 05 0 5 10 15 20 25 30 Time: s (b) 35 40 45 50

Test ID AA01 AA02

: degrees 28 28

ID : % Input motion (no.) 56 59 Chi-Chi, 1999 (4) Kobe, 1995 (4)

Peak input acceleration: g 0.41 0.43

Fig. 2. Input bedrock motions in time domain: (a) Chi-Chi (test AA01); (b) Kobe (test AA02)

AFTERSHOCKS AND THE WHOLE-LIFE SEISMIC PERFORMANCE OF GRANULAR SLOPES information to subsequently be collected for points far from the crest and toe of the slope (although these are not reported in this paper). A dynamic ground displacement was applied along the bottom edge of the nite-element (FE) model, shown by the repeating arrows in Fig. 1(b). The input motion applied to the model was that measured at instrument 8 in the centrifuge model: that is, the motion that the slope in the centrifuge actually saw, accounting for any losses between the shaking table and the container, and between the container bottom and the soil. The motions were input as ground displacements, determined by high-pass ltering and integration of the accelerometer records: ltering before integration to obtain velocity, and again before integrating velocity to obtain displacement, ensures that there is no permanent wander due to any offset in the accelerometer recordings or integration of random noise within the signal. Displacement data were extracted from the FE models at the locations of the instruments in the centrifuge tests shown in Fig. 1(a). At points 115, accelerations were subsequently determined from double differentiation of the displacements. CONSTITUTIVE MODELLING An elasto-plastic soil model with isotropic hardening (Schanz et al., 1999) is used in the modelling presented herein, in which the elastic behaviour incorporates straindependent stiffness variation following the model proposed by Hardin & Drnevich (1972), as modied by Santos & Correia (2001) G 1 G0 1 0:385js =s,0:7 j (1)

1233

where s is shear strain and s,0:7 is the shear strain at which G/G0 is 70%. Plastic failure is modelled using a cap-type yield surface combined with the MohrCoulomb failure criterion. This model is included in the PLAXIS FE suite used for the work described herein as the hardening soil model with small-strain stiffness (Benz, 2006). Although this model captures only strain-hardening (and not strainsoftening) behaviour, it will subsequently be shown that the effects of strain-softening on slope movement are minor when the earthquake is strong enough to induce signicant slip, as in the motions used in the centrifuge testing. The selection of appropriate soil strength (peak or critical state friction angles) will be discussed in greater detail later in this section. The model requires 13 input parameters: unit weights under saturated and dry conditions; three measurable effective stress strength parameters, p 9k , c9 and 9 (angle of dilation), for use in the MohrCoulomb failure criterion; six measurable stiffness parameters (which are stress dependent) describing the response to deviatoric loading (E50 ), compressive loading (Eoed ), unloadreload cycles (Eur , ur ), smallstrain stiffness (G0 ) and a shear strain for describing the shape of the G s relationship (s,0:7 ); and nally, two empirical parameters Rf and m, the former controlling the deviatoric stress at failure, and the latter controlling the variation of the stiffness parameters with effective conning stress.   c9 cos 9 3 9 sin 9 m E 3 9 Eref (2) c9 cos 9 pref sin 9   c9 cos 9 3 9 =K 0 sin 9 m Eoed 3 9 Eref (3) oed c9 cos 9 pref sin 9 where E may be E50 , Eur or G0 ; Eref is the value of the parameter at a reference stress of pref ; and K0 1 sin 9.

A reference pressure of pref 100 kPa is used throughout the remainder of this paper. Although the model has several input parameters, all except Rf can be measured in some form through routine laboratory testing. In the rst set of FEM simulations that will be described in the subsequent sections, the empirical correlations between the input parameters and relative density for coarsegrained soils presented by Brinkgreve et al. (2010) were used. Use of these parameters required no additional soil testing, and all the constitutive parameters can be dened using relative density, ID , alone (which was uniform within the centrifuge model slopes). Use of this model represented the case in practice where no detailed laboratory test data are available, but where relative density can be determined from standard penetration tests (SPTs) or a cone penetration test (CPT) prole (e.g. as outlined in Knappett & Craig, 2012). A second set of simulations was then conducted in which the key strength and stiffness parameters were calibrated for the soil used in the centrifuge tests using laboratory test data, the stiffness parameters being important in modelling the dynamic effects and shear-wave propagation, and the strength properties controlling the permanent deformations. This model is henceforth termed HST95. A large amount of published shearbox test data was available from previous studies that used the HST95 sand (Lauder, 2011; Bransby et al., 2011); this was used herein to derive soil-specic strength parameters, supplemented with oedometer tests to determine soil-specic stiffness parameters. In a practical case it would be necessary only to undertake tests appropriate for the in situ soil states (e.g. density in this case); however, in this paper a range of test data were collated/ obtained for reconstituted samples covering a wide range of relative densities such that a more complete model could be developed for use in future simulations of physical test results that use this material. Strength parameters Figure 4 shows a summary of shearbox data from a total of 38 tests conducted over a range of conning normal effective stresses between 5 kPa and 200 kPa, as summarised in Table 3. Fig. 4(a) shows shear stress measurements at critical state (when volumetric change had stopped), indicating that the critical state friction angle is c 9s 328. Fig. 4(b) shows the secant peak friction angles measured over the stress ranges considered (stress-independent values were determined by straight-line ts to the peak strength data, as p 9k is stress-independent within the constitutive model). A straight-line t to the data as a function of relative density appeared to be appropriate; focusing on the data points below ID 80% gave p 9k 20I D 29 (degrees) (4) Dilation angles are also shown in Fig. 4(b); the straightline t for this data was found to be 9 25I D 4 (degrees) (5) These simple linear ts to p 9k and 9 satisfy the dilatancy relationship given by Bolton (1986), and predict the value of c 9s 328 with an error of , 1% using either this relationship, or that by Rowe (1962), which is incorporated in the constitutive model formulation. Whereas the strength properties given by equations (4) and (5) match the element test data well, their use within a strain-hardening model would imply that the peak strength is appropriate to the analysis. It will subsequently be demonstrated that, for large-strain slope problems, the permanent deformations are governed by the critical state strength.

1234
140

AL-DEFAE, CAUCIS AND KNAPPETT

120 100 80 60 Lauder (2011) 40 20 0 0 50 100 150 200 Normal effective stress, n: kPa (a) Lauder (2011) Bransby et al. (2012) This paper 50 40 30 20 10 0 0 10 20 40 60 Relative density, ID: % (b) 80 100 Brinkgreve et al. (2010) 250 Radius 38 mm ( 2h) Bransby et al. (2012) This paper cs 32 h 19 mm C L Applied pressure

Shear stress at critical state, : kPa

70

Fig. 5. FE mesh used in simulating oedometer tests, showing boundary conditions

Peak friction angle, pk; dilation angle, : degrees

60

were determined for loose and dense samples, from which linear interpolations were made as a function of relative density, as presented by Brinkgreve et al. (2010). The interpolated parameters were subsequently checked against further tests at intermediate densities. Fig. 6(a) shows a comparison of some results for the Brinkgreve et al. (2010) model, which overpredicts stiffness in dense sand and dramatically underpredicts stiffness in loose sand. Fig. 6(b) shows the markedly improved results using tted values of Eoed , E50 , Eur and m. To reduce the number of independent parameters, E50 1.25Eoed and Eur 3Eoed were assumed,
Applied veritical effective stress: kPa
700 600 500 400 300 200 100 0 0 700 600 500 400 300 200 100 0 0 1 2 3 4 Vertical strain: % (b) 5 Oedometer, ID 20% FEM, ID 20% Oedometer, ID 73% FEM, ID 73% 1 2 3 Vertical strain: % (a) 4 5 Oedometer, ID 20% FEM, ID 20% Oedometer, ID 73% FEM, ID 73%

Fig. 4. DSA (shearbox) test data used in soil-specic calibration of constitutive model: (a) c 9s ; (b) p 9 k and 9

Hence, in modelling the centrifuge tests, p 9k c 9s 328 and 9 08 were used. Stiffness parameters Oedometer tests were conducted on dry samples of sand prepared by air pluviation within a Clockhouse Engineering Ltd J550 oedometer (sample height 19 mm; sample diameter 76 mm) at a range of relative densities, ID , between 5% and 83%. The tests were conducted up to an effective stress of 600 kPa, and included three unloadreload cycles (200100200 kPa; 400200400 kPa; 600400600 kPa) to ensure that Eur was well calibrated (this parameter is likely to be important during cyclic seismic loading). The constitutive model parameters were then determined by conducting virtual oedometer tests using FEM, starting with the parameters suggested by Brinkgreve et al. (2010), and modifying them as necessary to achieve a reasonable t to the data. Fig. 5 shows the axisymmetric model geometry and boundary conditions employed. To develop a complete density-dependent model in this paper, calibrated parameters
Table 3. DSA test data for HST95 silica sand Source Lauder (2011) Bransby et al. (2011) This paper (see Fig. 17) ID : % 17 40 75 9 41 93 55

Fig. 6. Comparison of one-dimensional compression curves for loose and dense samples: (a) using Brinkgreve et al. (2010) parameters; (b) using HST95 (soil-specic) parameters

No. of tests 8 6 4 5 5 5 5

Applied veritical effective stress: kPa

Effective normal stresses: kPa 5, 8, 11, 16, 30, 35, 70, 125 16, 30, 55, 100, 135, 150 11, 16, 30, 70 10, 25, 50, 100, 200 15, 35, 55, 100, 150 10, 25, 50, 100, 180 5, 8, 13, 16, 25

AFTERSHOCKS AND THE WHOLE-LIFE SEISMIC PERFORMANCE OF GRANULAR SLOPES so that Eoed could be used to simulate the strain magnitude correctly, and m used to control the shape of the stress strain curve. E50 Eoed was proposed by Brinkgreve et al. (2010), but this is not exactly true from comparison of equations (2) and (3). The value of 1.25 ensured that no unrealistic values of K0 would be implied in denser soils (including that used in the centrifuge tests); the analyses appeared to be relatively insensitive to this assumption at lower densities. Following iteration, the best-t stiffness parameters were determined to be (7)
G/G0
Chi-Chi (centrifuge data) Kobe (centrifuge data) Chi-Chi (FEM data) Kobe (FEM data) Hardin & Drnevich (1972) Ishibashi & Zhang (1993) Santos & Correia (2001) 10 08 06 04

1235

: Eref oed 25I D 20 22 (MPa) : : m 0 6 0 1I D

(6)

The power coefcient m is the inclination of stiffness stress curves normalised by pref in double logarithmic scale. The proposed calibration for m (equation (7)) gives a powerlaw exponent for stress dependence of stiffness that is between 0.5 and 0.6 at all densities. This is consistent with previous studies (e.g. Lo Presti et al., 1998), while maintaining the slight negative correlation between the two parameters noted by Brinkgreve et al. (2010). As neither of the aforementioned tests measures smallstrain parameters, G0 was estimated using the relationship based on void ratio (e) proposed by Hardin & Drnevich (1972), Gref 0 33 (2:97 e)2 (MPa) 1e (8)

02 0 0001

001 Shear strain: % (a)

01

05 04

Chi-Chi (centrifuge data) Kobe (centrifuge data) Chi-Chi (FEM data) Kobe (FEM data) Hardin & Drnevich (1972) Ishibashi & Zhang (1993) Santos & Correia (2001)

Damping

for pref 100 kPa. Equation (8) can be linearised, ignoring terms of order e 2 and above; expressing e in terms of relative density with the values of emax and emin gives : Gref 0 50I D 88 80 (MPa) (9)

03 02 01

The shear strain parameter s,0:7 was assumed to increase linearly from 0.01% at ID 20% to 0.02% at ID 80%: that is s,0:7 1:7I D 0:67 (310 )
4

0 0001

001 Shear strain: % (b)

01

(10)

Other parameters and comments Of the remaining parameters, the default value of Rf 0.9 was used, and the unit weight parameters were determined from standard relationships as a function of relative density and linearised, giving dry 3I D 14:5 (kN=m3 ) sat 1:8I D 18:8 (kN=m )
3

Fig. 7. Comparison of shear modulus degradation and damping in centrifuge tests and FE simulations, along with other published correlations

(11) (12)

The particular choice of direct shear apparatus (DSA) and oedometer tests for determining the parameters, as described above, was guided by available test data. Triaxial test data could equally be used to determine both strength and stiffness data, by simulating triaxial compression on a virtual test sample using FEM in a similar way to that described for the oedometer tests. Damping will be discussed later in the paper. VALIDATION OF FEM Dynamic shear modulus and damping Shear modulus and damping as functions of cyclic shear strain within both the centrifuge models and the numerical simulations are shown in Fig. 7. The data points representing the centrifuge models were determined from second-order estimates using the accelerometer data, following the method

proposed by Brennan et al. (2005). The data points for the numerical simulations were determined in the same way, using data from virtual accelerometers at homologous points within the FE mesh. Fig. 7 demonstrates that the operative shear modulus at a given cyclic shear strain is comparable to that measured in the centrifuge tests, but that the material hysteretic damping within the model is generally smaller than that inferred from the centrifuge tests. This will be revisited in the following section. Nonetheless, it would appear that the use of simple monotonic test data is effective in dening a constitutive model that can represent the principal dynamic behaviour of the soil. Accelerations Figure 8 shows a comparison of the measured and simulated acceleration at the toe of the slope (instrument 14) in the rst earthquake of test AA01 (Chi-Chi motion) in both the time and frequency domains (Figs 8(a) and 8(b) respectively). In this gure three cases are considered: (i) the use of Brinkgreve et al. (2010) constitutive parameters; (ii) the use of HST95 parameters; and (iii) the use of HST95 parameters with additional Rayleigh damping. The constitutive model implicitly includes material hysteretic damping

1236
Acceleration: g
Case (i)

AL-DEFAE, CAUCIS AND KNAPPETT

Acceleration: g
Centrifuge Brinkgreve et al. (2010) FEM

05 0 05
0

05 0 05
0

Case (i)

Centrifuge Brinkgreve et al. (2010) FEM

10

15

20

25

30

35

40

45

50

10

15

20

25

30

35

40

45

50

Acceleration: g

Acceleration: g

05 0 05
0

Case (ii)

Centrifuge HST95 silica sand FEM

05 0 05
0

Case (ii)

Centrifuge HST95 silica sand FEM

10

15

20

25

30

35

40

45

50

10

15

20

25

30

35

40

45

50

Acceleration: g

05 0 05
0

Case (iii)

Centrifuge HST95 sand Rayleigh FEM

Acceleration: g

05 0 05
0

Case (iii)

Centrifuge HST95 sand Rayleigh FEM

10

15

20

25 30 Time: s (a)

35

40

45

50

10

15

20

25 30 Time: s (a)

35

40

45

50

Magnitude: g/Hz

Magnitude: g/Hz

40 30 20 10 0 40 20 0

Case (i)

Centrifuge Brinkgreve et al. (2010) FEM

40 30 20 10 0

Centrifuge Brinkgreve et al. (2010) FEM Case (i)

1 Case (ii)

Magnitude: g/Hz

Magnitude: g/Hz

Centrifuge HST95 silica sand FEM

40 20 0

Centrifuge HST95 silica sand FEM Case (ii)

1 Case (iii)

Magnitude: g/Hz

40 20 0

Magnitude: g/Hz

Centrifuge HST95 sand Rayleigh FEM

40 20 0

Centrifuge HST95 sand Rayleigh FEM Case (iii)

2 3 Frequency: Hz (b)

Fig. 8. Comparison of measured and predicted accelerations at slope toe during test AA01: (a) time domain; (b) frequency domain

2 3 Frequency: Hz (b)

within in its formulation; however, these models generally overpredicted accelerations (as a result of underpredicting damping; see Fig. 7). The Rayleigh damping formulation allows additional mass and/or stiffness-proportional (modal, frequency-dependent) damping (add ) to be included   1 ck ( f n ) (13) add cm 4 f n where add is the additional equivalent viscous damping ratio, and fn is the natural frequency of modes within the soil. Values of cm 0.0005 and ck 0.005 were selected for use in case (iii), and the justication for these will be given later. At point 14 the match between the centrifuge test and FEM is good for all soil parameter sets considered, although the models without the additional Rayleigh damping slightly overpredict the peak acceleration values in some of the cycles. This match is not surprising, given that the shear wave has propagated only a short distance in level ground at this point, and has not yet interacted with the slope. Figure 9 shows similar plots at the crest of the slope (instrument 5), representing a more stringent test of the capabilities of the numerical model. This would be of sig-

Fig. 9. Comparison of measured and predicted accelerations at slope crest during test AA01: (a) time domain; (b) frequency domain

nicant interest for being able to determine the input motion to any (infra)structure located at the crest of the slope. For both cases (i) and (ii) without any additional damping, signicant overprediction of acceleration is observed. From Fig. 9(b) it is clear that this appears to arise from increased amplication of the higher-frequency components (above 3 Hz). The Rayleigh damping parameters reported earlier give damping that is predominantly stiffness-proportional, such that the motions at higher frequencies would be more signicantly damped, without over-damping the lower frequencies where the match is better. Case (iii) in Fig. 9(b) shows a markedly improved prediction of the acceleration time history at this point, owing to reduced contribution of the higher-frequency modes. Similar results were found for simulations of test AA02 (Kobe motion). Increases in the ck parameter, to increase and further improve the damping at higher frequencies, resulted in poorer predictions of permanent movements. The values for cm and ck given above appear to represent the best compromise, enabling both ground accelerations and permanent slip (see later) to be reasonably estimated within the same analysis. The ratios of crest peak acceleration (at instrument 5) to

AFTERSHOCKS AND THE WHOLE-LIFE SEISMIC PERFORMANCE OF GRANULAR SLOPES the peak value of the input motion were found to be insensitive to the number of earthquakes, and of the order of 1.3 (highest value recorded was 1.4). This amplication factor clearly contains two distinct effects: the site effect (in essence a material property effect, due to the dynamic properties of the soil); and a topographic effect (a geometric effect due to the ground surface prole). In Eurocode 8, Part 5 (BSI, 2005b), the ground motion at the crest (instrument 5), for use in constructing response spectra, is calculated using PGA S S T ag (14) where ag is the peak acceleration in the underlying bedrock (input motion in the centrifuge tests); S is the soil factor describing the site effect ( 1.4 for the ground type E soil in this study, as classied using Eurocode 8, Part 1 (BSI, 2005a)); and ST is the topographic amplication factor (> 1.2 for shallow slopes). The overall amplication factor is then PGA/ag : Based on Eurocode 8, the overall amplication within the centrifuge test would be predicted to be between 1.4 and 1.7, which would be conservative when compared with the centrifuge observations (amplication  1.3) for the two motions considered. Ashford et al. (1997) and Bouckovalas & Papadimitriou (2005) have conducted detailed numerical studies based on harmonic ground shaking, and have demonstrated that the topographic amplication is dependent on the ratio of slope height to wavelength and position from the crest. In light of this, the centrifuge tests of Brennan & Madabhushi (2009) are a useful comparison with the test data in this paper, as they considered a slope of similar (although not identical) height, slope angle and sand. Their results support the value of 1.3 measured in tests AA01 and AA02. However, the hazard posed to infrastructure at the crest is more usefully represented by a response spectrum, rather than a ratio of peak accelerations, which represents the spectral response at a period T 0 s only. Fig. 10 shows a

1237

comparison of predicted crest spectra for 5% nominal structural damping from FEM and measured data from the centrifuge tests, also including design spectra based on Eurocode 8 for context. For the cases including Rayleigh damping (HST95, case (iii)), a very good match to the centrifuge data is obtained, although there is a tendency for the response to be slightly underpredicted for periods above 0.5 s. Use of the other constitutive models results in a much more signicant underprediction of response above 0.5 s, and overprediction below this. Fig. 10 also suggests that there is a range of natural periods (between approximately 0.4 s and 1.0 s) over which the FEM may substantially underpredict the response, compared with the centrifuge data (this is particularly noticeable for test AA02). This range could be used as a simple screening tool to identify key pieces of infrastructure atop slopes that may be more vulnerable to seismic damage than FEM would suggest, and to which extra consideration should be paid in design. It is clear that care should be taken in interpreting the hazard to supported infrastructure from analyses using FEM, and that this will be signicantly dependent on the natural period of the supported structure. Further comparison between the rst and last earthquake motions demonstrated that the spectral magnitudes appear to be relatively insensitive to repeated strong earthquake shaking. Permanent deformations Figure 11 shows the permanent crest settlements across the four earthquakes as predicted by FEM (all three cases are shown), and as measured in the centrifuge. It is clear that, for both suites of earthquake motions, use of the Brinkgreve et al. (2010) parameters hugely overpredicts settlement at the crest. This would lead to signicant overprediction of the risk posed to the slope, and hence, potentially, uneconomic design. In contrast, use of the HST95 parameters (either case) gives a much better prediction, although inclusion of the additional damping, which was benecial for modelling the dynamic behaviour accurately, results in an underprediction of permanent deformation (which is unsafe). The Brinkgreve et al. (2010) parameters were initially envisaged as being useful for cases where there is extremely limited site investigation data, but the implication of Fig. 11 is that it is always important to obtain soil parameters from high-quality laboratory (or in situ) test data to achieve an accurate prediction of movement. It is likely that the extra cost of this additional investigation would be signicantly offset if a lower amount of remediation/repair were to be required (as a result of a more accurate prediction of seismic response). When considered alongside the dynamic performance discussed in the previous section, it is clear that neither case (ii) or case (iii) can give consistently better performance of both dynamic and permanent movements simultaneously; this would appear to suggest that the additional Rayleigh damping may not be a material characteristic, but may be masking an effect of the sloping ground geometry in which wave reection at the sloping ground surface is not modelled correctly within the FE model. Figure 12 shows a comparison (drawn to scale) of the ground surface prole measured at the end of the centrifuge test (a negligible amount of movement was recorded during spin-down), and as predicted at the end of the last earthquake (EQ4) from the FE model (case (ii)) for test AA01. It can be seen that the numerical model captures the deformed shape of the slope well, particularly the angular distortion at the crest, which is likely to be of greatest signicance for supported infrastructure. A similar result was obtained for test AA02 (not shown).

35

Spectral acceleration: g

30 25 20 15 10 05 0 0 02 04 06 08

Centrifuge (AA01) Brinkgreve et al. (2010) FEM HST95 sand FEM HST95 sand FEM Rayleigh EC8 (S 14, ST 12)

10 12 Period: s (a)

14

16

18

20

35

Spectral acceleration: g

30 25 20 15 10 05 0 0 02 04 06 08

Centrifuge (AA02) Brinkgreve et al. (2010) FEM HST95 sand FEM HST95 sand FEM Rayleigh EC8 (S 14, ST 12)

10 12 Period: s (b)

14

16

18

20

Fig. 10. Measured (centrifuge), predicted (FEM) and design (Eurocode 8) response spectra at top of slope (instrument 5) for 5% structural damping: (a) Chi-Chi (AA01); (b) Kobe (AA02)

1238
0 05 10 15

AL-DEFAE, CAUCIS AND KNAPPETT

Settlement: m

20 25 30 35 40 45 50 55 0 Centrifuge settlement (091 m) Brinkgreve et al. (2010) FEM (505 m) HST95 sand FEM (079 m) HST95 sand FEM Rayleigh (041 m) 50 100 Time: s (a) 150 200

0 05 10

Settlement: m

15 20 25 30 35 0 Centrifuge settlement (102 m) Brinkgreve et al. (2010) FEM (33 m) HST95 sand FEM (076 m) HST95 sand FEM Rayleigh (032 m) 50 100 Time: s (b) 150 200

Fig. 11. Comparison of permanent crest settlements from FEM and centrifuge modelling: (a) test AA01; (b) test AA02
079 m (HST95 FEM)
khg After centrifuge g L

Original profile

091 m (centrifuge)

khzL cos

8 (160)

After FEM

ultL ( u)L

14 (280)

(Centrifuge) 195 m

6 (120)

zL cos

Fig. 13. Forces acting within innite slope under horizontal shaking

11 (220)

15 (300)

75 (150)

Fig. 12. Comparison of slope prole after EQ4 as predicted by FEM and as measured in centrifuge, test AA01

uniaxial horizontal shaking (plane strain), the applied downslope shear stress is applied z sin cos k h z cos2 (15)

DEVELOPMENT OF AN IMPROVED SLIDING-BLOCK METHOD The horizontal yield acceleration of a shallow translational (innite) slip can be determined using standard limit equilibrium techniques, incorporating pseudo-static acceleration components due to the seismic ground motion (see Fig. 13). For a slip plane at depth z beneath the slope surface under

where the rst term relates to the static shear stress due to the ground slope, and the second term relates to the additional peak dynamic shear stress induced by the earthquake. The shear strength of the soil along the slip plane, assuming that the soil failure can be described by the MohrCoulomb failure criterion, is given by

AFTERSHOCKS AND THE WHOLE-LIFE SEISMIC PERFORMANCE OF GRANULAR SLOPES ult c9 9 tan 9 c9 (z cos2 k h z sin cos u) tan 9 (16)

1239

the slip plane (so long as it continues to be parallel to the slope surface). Equation (17) then simplies to k hy cos tan 9 sin cos sin tan 9 (18)

The soil yields when applied ult : The value of kh at which this occurs (the yield acceleration, khy ) can be determined from equations (15) and (16) as k hy c9 (z cos u) tan 9 z sin cos z cos2 z sin cos tan 9
2

(17)

In a standard Newmark analysis, when the horizontal component of the ground acceleration (a) exceeds khy in the downslope direction, the slope will start to accelerate under the slip acceleration, aslip a khy (positive value implies downslope movement). This acceleration is numerically integrated with respect to time to obtain the slip velocity, which is then itself integrated to obtain the slip displacement. Once a , khy , the sliding block will begin to decelerate (as aslip , 0) until the slip velocity reaches zero, at which point the block comes to rest until aslip is again positive. This procedure is shown schematically in Fig. 14(a). In a dry cohesionless soil, c9 0, u 0, and z cancels in equation (17): that is, khy is independent of the depth of
Including strain-softening (SS): khy f () Classical Newmark: khy const.

Time

Ground motion

Classical Including SS

Time

In equation (18) the only parameters affecting khy are the slope angle (geometric) and the soil friction angle 9 (constitutive). In a standard analysis, both and 9 are constant. In reality, however, the soil may be strain-softening, in which case 9 will depend on the magnitude of the shear strain (s ) on the slip plane and the density of the soil. Matasovic et al. (1997) presented a simplied model for this, 9s , the which is shown schematically in Fig. 14(b). If p 9k c model reduces to the standard case of a strain-softening material. To incorporate this into an analysis, the value of khy at a particular time step can be estimated, based on the current permanent downslope displacement, computed in the previous time step, divided by the thickness of the shear band/slip plane to obtain an estimate of the shear strain. The slope angle will also change during an analysis, as slip will cause settlement at the crest and accumulation of material at the toe: that is, the slope will become shallower (re-grading, RG). A simplied model for this geometric effect is developed in this paper, and is shown in Fig. 15. Fig. 15(a) shows the kinematically admissible failure mechanism assumed for an increment of slip, d i , in which innite sliding is the predominant component. This leads to downward vertical movement of the material at the slope crest and a horizontal translation of the position of the toe, resulting in a new, shallower slope with an approximate slope angle of i1 in Fig. 15(b). Ambrayses & Srbulov (1995) present an alternative mechanism for shallow sliding that is similar to Fig. 15(a), but with more complex changes in geometry (and for post-seismic sliding only, i.e. when there is no seismic inertia). The new (simpler) mechanism proposed here has the advantage that the calculations are much more straightforward, relying on only a single parameter (the initial slope angle ) to dene the slope geometry, and more closely match the deformations observed visually in the centrifuge tests (e.g. see Fig. 12). Provided that is relatively small (such that the slope is long compared with its height), the equilibrium of this mechanism will be well approximated by innite slope theory (i.e. equations (17) or (18) will adequately describe
Crest Settlement increment di sin i Slip increment, di Toe

Slip displacement

Slip velocity

Acceleration

Time (a) Including strain-softening (SS): khy f () di cos i (a) New slope surface (i 1)

pk

khy and

cs Hi

Classic Newmark: khy const.

Shearbox test curve tan1 (/) vs s

Hi di sin i i Hi cot i (b)

i1

s,pk

s,cs (b)

Shear strain, s

di cos i

Fig. 14. Newmark sliding-block procedure, and effect of strainsoftening (after Matasovic et al., 1997)

Fig. 15. (a) New incremental slope re-grading mechanism; (b) incremental changes in geometry

1240

AL-DEFAE, CAUCIS AND KNAPPETT

khy ). It will be demonstrated later that even a slope as steep as ,1:2 satises this condition to a reasonable degree of accuracy. It is assumed that any volumetric change in the material in the sliding block is negligible. From Fig. 15(b), the instantaneous slope angle i1 can be determined from   H i d i sin i 1 (19) i1 tan H i cot i d i cos i where Hi is the height of the slope at the previous time step. For the initial time step, d0 0, Hi H and i 0 (initial slope angle). It should be noted that the deformations in Fig. 15(b) are shown at exaggerated scale for clarity. The slope angle can therefore be recalculated at each time step to account for the re-grading of the slope based on the increment of slip occurring in the previous time step, as 9 was previously to account for strain-softening. The yield acceleration (equation (18)) will thus be fully strain-dependent. This method assumes that once the slope has been deformed to a new, smaller value of , the failure mechanism will continue to be of the innite type, with a new slip surface forming parallel to the new slope surface. It also assumes that the strain-dependent effects on and 9 are independent, to simplify the calculations. In reality, this latter assumption may not be wholly true, as the changing angle of a softened slip plane (i.e. with 9 c 9s ) may push at least part of the slip surface into previously undisturbed soil. If this effect and the effects of strain-softening are signicant, it is expected that the model will overestimate movements; however, this would be conservative for use in analysis and design. It should also be noted that the model as formulated can be used even for the case of large total slope movements (such as may accrue during a series of strong aftershocks), as the displacement increment in each individual time step will remain small, and therefore the instantaneous failure mechanism will be represented by Fig. 15(a) for small displacement increments. From the form of equation (19) it is clear that can only ever reduce during an earthquake, resulting in an increase of khy from equation (17) or (18): that is, the slope will geometrically harden during an earthquake, and the slip in a subsequent (identical) earthquake will be less than that occurring in the original. The behaviour of a seismically damaged slope during a subsequent earthquake can therefore be determined by starting from the initial conditions (amount of slip, accumulated strain, re-graded slope angle and current friction angle) obtained at the end of the analysis for the previous ground motion. VALIDATION OF SLIDING-BLOCK MODEL Predictions of permanent deformation Visual observations from the centrifuge tests suggested that the 1:2 slopes tested failed in a shallow translational mechanism consistent with that shown in Fig. 15. The depth of the shear plane ( 0.5 m) was estimated using the discontinuity layout optimisation (DLO) technique (Smith & Gilbert, 2007) to obtain a minimum upper-bound mechanism for the actual limited geometry of the centrifuge model. DLO calculations were carried out using LimitState:GEO, v2.0. The position of the slip plane was not affected by the value of 9 used, and is shown in Fig. 16(a). The idealised geometry assumed in the analysis (cf. Fig. 15(a)) is shown in Fig. 16(b), overlaid onto the accumulated shear strain within the FE simulation of test AA01 (case (ii)), for comparison. Shearbox testing was then conducted on samples of dry sand prepared to the same relative density as in the centrifuge tests, using a standard 60 mm 3 60 mm DSA, 9s : These tests were conducted at a range to obtain p 9k and c

05 m

0 12345678 (a) Shear strain: % 400 360 320 280 240 200 160 120 80 40 0 (b)

Fig. 16. (a) Failure mechanism computed from DLO for seismic case; (b) accumulated shear strain within FE model (test AA01) and assumed approximate innite slip mechanisms used in Newmark analyses

of effective conning stresses representing those within the top 1 m of the soil. The test data are shown in Fig. 17, from 9s 328, s;pk 3.5% and which values of p 9k 408, c s;cs 7.5% were estimated for 0.5 m depth. Using these friction angles and limiting shear strains, the initial yield accelerations of the slope were computed using equation (18) and DLO; the results are shown in Table 4, alongside values for the static factor of safety of the slope (F), calculated using equations (15) and (16) with kh 0, and DLO. The shear band thickness (required for converting slip displacement in the sliding-block model into an approximate shear strain) was estimated at t 16D50 2.4 mm, based on a range of previous studies (e.g. Mu lhaus & Vardoulakis, 1987; Oda & Kazama, 1998; Muir Wood, 2002). The calculations were conducted at prototype scale, and so the shear strain was estimated using 50t 120 mm to model the correct ratio between the slope geometry and the grain size within the model. For application to a true eld case where the grains are smaller compared with the overall size of the slope, the true shear band thickness should be used instead. For the tests presented herein, changing the shear band thickness from 120 mm to 2.4 mm resulted in less than 1% change in crest settlement (the actual value varied slightly with the input motion considered), conrming that the grain10

08

Stress ratio ( tan )

06 55 kPa (034 m) 82 kPa (051 m) 02 136 kPa (085 m) Analytical model 0 0 5 10 Shear strain, s: % 15 20

04

Fig. 17. Soil test data from direct shear apparatus (DSA) at low effective conning stress

AFTERSHOCKS AND THE WHOLE-LIFE SEISMIC PERFORMANCE OF GRANULAR SLOPES

Table 4. Static and dynamic slope stability data Soil strength Static stability parameters F (equations (15) and (16)) 9 328 9 408 1.17 1.56 F (DLO) 1.20 1.61 Dynamic stability parameters khy (equation (18)) 0.07g 0.21g khy (DLO) 0.07g 0.22g

1241

Crest settlement: m

size scaling effect is negligible, and that centrifuge modelling is therefore an appropriate technique for modelling slope failure problems in coarse-grained soil. A similar conclusion was reached by Anastasopoulos et al. (2007) for fault rupture (shear band) propagation through the same sand. Figure 18 shows the effect of the geometric re-grading (change in ), using the rst earthquake (EQ1) of test AA01 as an example. Only the positive (downslope) accelerations have been shown, for clarity. No slip occurs until the ground motion exceeds the yield acceleration based on the peak strength. Once the ground motion exceeds this value, however, and the slope begins to slip, the shear strain rapidly accrues, resulting in softening to critical state conditions after the rst large pulse. Motion of the slope also causes re-grading (geometric hardening), and the yield acceleration can subsequently be seen to increase non-linearly throughout the remainder of the earthquake, leading to reduced slip velocities (and hence reduced permanent slip) compared with the case with no geometric hardening. Sliding-block analyses were subsequently conducted for each of the centrifuge tests (for all four earthquakes in the cases of tests AA01 and AA02). Simulations were conducted using the bedrock input motion (this was taken from the bottom-most accelerometer in the model, instrument 8), as used in the FEM. Additional simulations using instrument 6 did not show signicant differences in the overall cumulative slip predicted, possibly as there was some shear decoupling across/around the shear band, which counteracted (at least partially) any increase in acceleration due to soil amplication. In slopes with deep rotational failure surfaces (e.g. in municipal solid waste or steep cohesive slopes), the dynamic
040

behaviour of the material within the sliding block (due to its thickness) may be signicant (e.g. Rathje & Bray, 1999); this was not the case for the extremely shallow translational slips that occurred within the cohesionless slopes tested herein. Figures 19 and 20 show the results of simulations of cumulative crest displacement both with (SS + RG) and without re-grading (SS) with the centrifuge test data, for tests AA01 and AA02 respectively. It can be seen that, in each case, the improved model presented in this paper (SS + RG) tracks the settlement at the crest of the slope much more closely than the model that incorporates only the constitutive effect (SS). These latter models increasingly diverge from the measured values with further strong shaking, as they always start with the initial (steeper) slope geometry, and therefore overpredict the slip. If the input motions were identical, the slip in each subsequent earthquake would be identical for the case of no re-grading (although the movements in the rst earthquake might be slightly smaller, owing to the strain-softening effect). The improved models are not perfect, and in each case overpredict the measured movement; as the slope re-grades, the new position of the slip plane may cause it to pass at least

0 0

50

Time: s 100

150

200

2 Centrifuge (AA01) Sliding block (SS RG) Sliding block (SS) EQ1 4 EQ2 EQ3 EQ4

Acceleration: g

030 020 010

Strainsoftening Geometric hardening (re-grading)

Input ground motion khy SS RG khy SS

05

Input acceleration: g

021

016

020

023

025

10

20 Time: s (a)

30

40

50

0 0 50 100 150 200

10

Slip velocity: m/s

08 06 04 02 0

Strain-softening and re-grading (SS RG) Strain-softening only (SS)

05 Ground acceleration (AA01) khy (SS RG) khy (SS)

10

20 Time: s (b)

30

40

50

Fig. 18. Application of new sliding-block model, showing key features (Chi-Chi EQ1 used, test AA01)

Fig. 19. Comparison of predicted cumulative crest settlements (with and without re-grading) with centrifuge test measurements: test AA01 (Chi-Chi)

1242
0 0 50 Time: s 100

AL-DEFAE, CAUCIS AND KNAPPETT

150 200

2 Centrifuge (AA02) Sliding block (SS RG) Sliding block (SS) EQ1 4 EQ2 EQ3 EQ4

05 021 012 0 0 50 100 150 200 016

018

020

05 Ground acceleration (AA02) khy (SS RG) khy (SS)

strain-dependent sliding-block model (strain-softening and geometric hardening) and the laboratory-test-calibrated FEM developed in this paper (HST95, case (ii)). Although the improved models are not perfect, they give a much improved prediction of the response under the initial earthquake on virgin soil, and are subsequently able to give a better prediction of behaviour, even after several previous strong earthquakes during which signicant deformation has accrued. Both existing models overpredict settlements in the rst earthquake and then get progressively worse with further shaking, as they are unable to correctly capture the effect of the previous seismic (strain) history on the slope. In the case of the sliding-block model this is associated with incorrect description of the deformed slope; in the case of the FEM, the case (i) constitutive parameters do not appear to correctly capture the strain history or mechanical response of the soil. The development of the improved sliding-block and FEM tools in this paper provides an improved means of quantifying the response of shallow cohesionless slopes under strong earthquake shaking, and the ability to consider behaviour under multiple successive earthquakes. With further development, these tools will allow civil engineers to obtain a better estimate of the hazard associated with aftershocks, and lead to new approaches to quantifying seismic performance and managing critical transport infrastructure, in which whole-life performance can be considered. CONCLUSIONS In this paper, improved procedures for modelling the seismic response of dry granular shallow slopes using the FEM and a fully strain-dependent Newmark sliding-block procedure have been developed and validated against dynamic centrifuge test data. In the FE modelling, a nonlinear elasto-plastic constitutive model was used, the parameters of which could be estimated based only on relative density using existing correlations, or using routine laboratory tests to develop a soil-specic model. These methods, as they do not require specialist testing, offer signicant benets for use in routine design, although the predictions obtained are of an approximate nature. It was demonstrated that the use of soil-specic parameters gave far improved predictions, particularly of permanent settlement, compared with the existing correlations, which overpredicted settlements, particularly in subsequent earthquakes. This highlights the value of performing adequate site investigation. In the improved sliding-block model, reduction in slope angle with slip/strain (re-grading or geometric hardening) has been incorporated alongside an existing strain-softening/ hardening formulation. The sliding-block and FE models gave comparable predictions of permanent slip, capturing the signicant decay in ground displacement (geometric hardening) with subsequent shaking observed in the centrifuge tests. They therefore permit both the response of slopes under strong aftershocks, and the whole-life performance of a slope, to be quantied. Whereas the sliding-block models are useful in preliminary design, owing to the limited soil property data required and the reduced computational effort, the FEM is additionally able to quantify the dynamic performance of the soil and the ground deformation prole (angular distortion) at the crest. This would provide the information necessary to make a detailed study of the seismic hazard posed to infrastructure located at the slope crest, without requiring an excessive amount of specialist laboratory testing, and is therefore complementary to the sliding-block models, being useful in the later stages of detailed design. The FE models generally overpredicted the magnitude of dynamic ground motions in

Fig. 20. Comparison of predicted cumulative crest settlements (with and without re-grading) with centrifuge test measurements: test AA02 (Kobe)

partially through undisturbed soil, thereby leading to an enhanced average frictional resistance along the slip plane. COMPARISON OF SLIDING-BLOCK AND FEM Figure 21 presents a summary of the predictive ability of the FEM and the sliding-block models for determining permanent settlements. These have been grouped into existing models, representing the current state-of-the-art, namely sliding block with strain-softening and FEM using the previously published parameters of Brinkgreve et al. (2010), and improved models, consisting of the fully
12

Input acceleration: g

Crest settlement: m

Predicted/measured (crest settlement)

10

Improved methods: Sliding block (SS RG) FEM (ii) Existing methods: Sliding block (SS) FEM (i)

Parity

0 1 2 3 Earthquake no. 4

Fig. 21. Accuracy of existing models/procedures, compared with those proposed in this and the companion paper (improved models) for predicting permanent crest settlement

AFTERSHOCKS AND THE WHOLE-LIFE SEISMIC PERFORMANCE OF GRANULAR SLOPES the slopes (particularly for components above 3 Hz). Adding in additional damping using the Rayleigh formulation improved this, but had an adverse effect on the prediction of permanent slope movements.
p 9k 9 (secant) peak angle of friction effective angle of dilation

1243

ACKNOWLEDGEMENTS The authors would like to express their sincere gratitude to Mark Truswell and Colin Stark at the University of Dundee for their assistance in performing the centrifuge tests. The rst author would also like to acknowledge the nancial support of his PhD studies from the Ministry of Higher Education and Scientic Research (MOHESR) of the Republic of Iraq.

REFERENCES
Ambrayses, N. & Srbulov, M. (1995). Earthquake induced displacements of slopes. Soil Dynam. Earthquake Engng 14, No. 1, 5971. Anastasopoulos, I., Gazetas, G., Bransby, M., Davies, M. & El Nahas, A. (2007). Fault rupture propagation through sand: niteelement analysis and validation through centrifuge experiments. J. Geotech. Geoenviron. Engng 133, No. 8, 943958. Ashford, S. A., Sitar, N., Lysmer, J. & Deng, N. (1997). Topographic effects on the seismic response of steep slopes. Bull. Seismol. Soc. Am. 87, No. 3, 701709. Benz, T. (2006). Small-strain stiffness of soils and its numerical consequences. PhD thesis, University of Stuttgart, Germany. Bertalot, D. (2012). Behaviour of shallow foundations on layered soil deposits containing loose saturated sands during earthquakes. PhD thesis, University of Dundee, UK. Bertalot, D. & Brennan, A. J. (2012). Inuence of bearing pressure on liquefaction-induced settlement of shallow foundations. otechnique 63, No. 5, 391399, http://dx.doi.org/10.1680/ Ge geot.11.P.040. Bertalot, D., Brennan, A. J., Knappett, J. A., Muir Wood, D. & Villalobos, F. A. (2012). Use of centrifuge modelling to improve lessons learned from earthquake case histories. Proceedings of the 2nd European conference on physical modelling in geotechnics, Eurofuge 2012, Delft, the Netherlands. Bolton, M. D. (1986). The strength and dilatancy of sands. otechnique 36, No. 1, 6578, http://dx.doi.org/10.1680/geot. Ge 1986.36.1.65. Bouckovalas, G. D. & Papadimitriou, A. G. (2005). Numerical evaluation of slope topography effects on seismic ground motion. Soil Dynam. Earthquake Engng 25, No. 710, 547558. Bransby, M. F., Brown, M. J., Knappett, J. A., Hudacsek, P., Morgan, N., Cathie, D., Maconochie, A., Yun, G., Ripley, A. G., Brown, N. & Egborge, R. (2011). Vertical capacity of grillage foundations in sand. Can. Geotech. J. 48, No. 8, 12461265. Brennan, A. J. & Madabhushi, S. P. G. (2009). Amplication of seismic accelerations at slope crests. Can. Geotech. J. 46, No. 5, 585594. Brennan, A. J., Thusyanthan, N. I. & Madabhushi, S. P. G. (2005). Evaluation of shear modulus and damping in dynamic centrifuge tests. J. Geotech. Geoenviron. Engng 131, No. 12, 14881497. Brinkgreve, R. B. J., Engin, E. & Engin, H. K. (2010). Validation of empirical formulas to derive model parameters for sands. In Numerical methods in geotechnical engineering (eds T. Benz and S. Nordal), pp. 137142. Rotterdam, the Netherlands: CRC Press/Balkema. BSI (1990). BS 1377, Part 4: Maximum and minimum dry density for granular soil. London, UK: British Standards Institution. BSI (2005a). EN 1998-1: 2004: Eurocode 8: Design of structures for earthquake resistance Part 1: General rules, seismic actions and rules for buildings. London, UK: British Standards Institution. BSI (2005b). EN 1998-5: 2004: Eurocode 8: Design of structures for earthquake resistance Part 5: Foundations, retaining structures and geotechnical aspects. London, UK: British Standards Institution. Hardin, B. O. & Drnevich, V. P. (1972). Shear modulus and damping in soils: design equations and curves. J. Soil Mech. Found. Div., ASCE 98, No. SM7, 667692. Ishibashi, I. & Zhang, X. (1993). Unied dynamic shear moduli and damping ratios of sand and clay. Soils Found. 33, No. 1, 182191. Jibson, R. W. (1993). Predicting earthquake-induced landslide displacements using Newmarks sliding block analysis. Transp. Res. Record 1411, 917. Khazai, B. & Sitar, N. (2004). Evaluation of factors controlling earthquake-induced landslides caused by the Chi-Chi earthquake and comparison with the Northridge and Loma Prieta events. J. Engng Geol. 71, No. 12, 7995. Kim, J. & Sitar, N. (2004). Direct estimation of yield acceleration

NOTATION
a ground acceleration ag underlying bedrock peak acceleration aslip slip acceleration Cu coefcient of uniformity C2 coefcient of curvature c9 cohesion intercept ck stiffness-proportional Rayleigh damping coefcient cm mass-proportional Rayleigh damping coefcient D10 particle diameter at which 10% is smaller D30 particle diameter at which 30% is smaller D60 particle diameter at which 60% is smaller d slope parallel slip Eoed oedometric tangent stiffness (in compression) Eur unloadingreloading stiffness E50 triaxial secant stiffness (at 50% of deviatoric failure stress in drained triaxial compression) e natural void ratio emax maximum void ratio emin minimum void ratio F static factor of safety fn natural frequency G shear modulus G0 small-strain modulus Gs specic gravity g acceleration due to gravity ( 9.81 m/s2 ) H slope height above toe ID relative density K0 lateral earth pressure coefcient (at rest) kh pseudo-static seismic horizontal acceleration khy yield acceleration Mw moment magnitude m power-law index for stress-level dependence of stiffness PGA peak ground acceleration (at soil surface) pref reference stress ( 100 kPa) Rf ratio of deviatoric failure stress to asymptotic limiting deviator stress S soil factor (Eurocode 8) ST topographic amplication factor (Eurocode 8) t shear band thickness u pore water pressure Vs shear wave velocity z depth of slip plane slope angle 0 initial slope angle (pre-earthquake) soil unit weight d dry unit weight sat saturated unit weight s shear strain s,cs shear strain at critical state s,pk shear strain at peak state add viscous damping ratio ur Poissons ratio (unloadreload) 9 normal effective stress applied applied shear stress ult shear strength 9 effective angle of friction c 9s critical state angle of friction

1244

AL-DEFAE, CAUCIS AND KNAPPETT

Newmark, N. M. (1965). Effects of earthquakes on dams and otechnique 15, No. 2, 139159, http://dx.doi. embankments. Ge org/10.1680/geot.1965.15.2.139. Oda, M. & Kazama, H. (1998). Microstructure of shear bands and its relation to the mechanism of dilatancy and failure of dense otechnique 48, No. 4, 465482, http://dx.doi. granular soils. Ge org/10.1680/geot.1998.48.4.465. Rathje, E. M. & Bray, J. D. (1999). An examination of simplied earthquake-induced displacement procedures for earth structures. Can. Geotech. J. 36, No. 1, 7287. Rowe, P. W. (1962). The stressdilatancy relation for static equilibrium of an assembly of particles in contact. Proc. R. Soc. A 269, No. 1339, 500527. Santos, J. A. & Correia, A. G. (2001). Reference threshold shear strain of soil: its application to obtain a unique strain-dependent shear modulus curve for soil. Proceedings of the 15th international conference on soil mechanics and geotechnical engineering, Istanbul, Turkey, vol. 1, pp. 267270. Schanz, T., Vermeer, P.A. & Bonnier, P.G. (1999). The hardeningsoil model: formulation and verication. In Beyond 2000 in computation geotechnics (ed. R. B. J. Brinkgreve), pp. 281290. Rotterdam, the Netherlands: Balkema. Smith, C. C. & Gilbert, M. (2007). Application of discontinuity layout optimization to plane plasticity problems. Proc. R. Soc. A 463, No. 2086, 24612484. Wartman, J., Bray, J. & Seed, R. (2003). Inclined plane studies of the Newmark sliding block procedure. J. Geotech. Geoenviron. Engng 129, No. 8, 673684. Wartman, J., Seed, R. & Bray, J. (2005). Shaking table modeling of seismically induced deformations in slopes. J. Geotech. Geoenviron. Engng 131, No. 5, 610622.

in slope stability analyses. J. Geotech. Geoenviron. Engng 130, No. 1, 111115. Knappett, J. A. & Craig, R. F. (2012). Craigs soil mechanics, 8th edn. London, UK: Spon Press. Kramer, S. L. (2008). Performance-based earthquake engineering: opportunities and implications for geotechnical engineering practice. Keynote paper. Proceedings of the 4th conference on geotechnical earthquake engineering and soil dynamics, pp. 1 32. Reston, VA, USA: ASCE. Lauder, K. (2011). The performance of pipeline ploughs. PhD thesis, University of Dundee, UK. Lo Presti, D. C. F., Pallara, O., Fioravante, V. & Jamiolkowski, M. (1998). Assessment of quasi-linear models for sands. In Prefailure deformation behaviour of geomaterials (eds R. Jardine, M. Davies, D. Hight, A. Smith and S. Stallebrass), pp. 363372. London, UK: Thomas Telford. Loukidis, D., Bandini, P. & Salgado, R. (2003). Stability of otechnique 53, seismically loaded slopes using limit analysis. Ge No. 5, 463479, http://dx.doi.org/10.1680/geot.2003.53.5.463. Lysmer, J. & Kuhlmeyer, R. L. (1969). Finite dynamic model for innite media. Proc. ASCE, J. Engng Mech. Div. 95, No. 4, 859887. Matasovic, N., Kavazanjian, E. Jr & Yan, L. (1997). Newmark deformation analysis degrading yield acceleration. Proceedings of Geosynthetics 97, Long Beach, CA, USA, vol. 2, pp. 989 1000. Muir Wood, D. (2002). Some observations of volumetric instabilities in soils. Int. J. Solids Struct. 39, No. 1314, 34293449. Mu lhaus, H. B. & Vardoulakis, I. (1987). The thickness of shear otechnique 37, No. 3, 271283, bands in granular soils. Ge http://dx.doi.org/10.1680/geot.1987.37.3.271.