canlex 4 modelling cgj

CANLEX full-scale experiment and modelling

P.M. Byrne, H. Puebla, D.H. Chan, A. Soroush, N.R. Morgenstern, D.C. Cathro,W.H. Gu, R. Phillips, P.K. Robertson, B.A. Hofmann, C.E. (Fear) Wride, D.C. Sego,H.D. Plewes, B.R. List, and S. Tan

Abstract: A major aim of the Canadian Liquefaction Experiment (CANLEX) was to verify analysis procedures for pre-dicting liquefaction phenomena. Towards this purpose, two loading events were carried out: a field event comprising aclay embankment built over a loose sand foundation layer, and a centrifuge test performed on a model of a sand em-bankment structure. Both the field event and the centrifuge model were planned so as to induce a static liquefactionfailure and were instrumented to observe their response in terms of displacement and pore pressure. The fundamentalmechanical characteristics of the foundation layer were determined from laboratory element tests (triaxial and simpleshear). These tests formed the basis for the stressstrain modelling used in the numerical analyses. Two fundamentallydifferent modelling techniques were used. One involved a fully coupled plasticity model, and the other involved amodel based on a collapse-surface approach. The model and prototype structures were then analyzed and the predictedresults in terms of displacements and pore pressures were compared with the measured values. The results from bothapproaches were found to be in reasonable agreement with the measurements, provided allowance was made for direc-tion of loading and drainage effects were accounted for.

Key words: liquefaction, field experiment, embankment, centrifuge model, elasticplastic model.

Rsum : Un but principal du Canadian Liquefaction Experiment (CANLEX) tait de vrifier les procdures danalysepour prdire les phnomnes de liqufaction. cette fin, deux cas de chargement ont t raliss: un chargement sur leterrain comprenant la construction dun remblai dargile sur une couche de fondation de sable meuble, et un essai aucentrifuge ralis sur un modle dune structure de remblai de sable. Tant le remblai sur le terrain que le modle dansle centrifuge ont t planifis de faon induire une rupture par liqufaction statique et ont t instruments pour ob-server leurs rponses en termes de dplacement et de pression interstitielle. Les caractristiques mcaniquesfondamentales de la couche de fondation ont t dtermines en partant dessais sur des spcimens en laboratoire(triaxiaux et cisailllement direct). Ces essais formaient la base pour la modlisation contraintedformation utilise danslanalyse numrique. Deux techniques de modlisation fondamentalement diffrentes ont t utilises. Une impliquaitun modle de plasticit compltement coupl, et lautre comprenait un modle bas sur une approche de surfacedeffondrement. Les structures du modle et du prototype ont alors t analyses et les rsultats prdits en termes dedplacements et de pressions interstitielles ont t compars avec les valeurs mesures. On a trouv que les rsultats deces deux approches taient en concordance raisonnable avec les mesures, condition que lon tienne compte des effetsde la direction du chargement et du drainage.

Mots cls : liqufaction, exprience sur le terrain, remblai, modle centrifuge, modle lastoplastique.

[Traduit par la Rdaction] Byrne et al. 562

Introduction

The Canadian Liquefaction Experiment (CANLEX) includedthree major aspects: characterization of sand to evaluate itsliquefaction characteristics; development and calibration ofnumerical models to predict liquefaction response; and afull-scale liquefaction event at the Syncrude site near FortMcMurray, Alberta, to validate predictive ability.

Liquefaction characterization was carried out at a numberof sites. This involved in situ testing as described by Wrideet al. (2000) and laboratory testing on both undisturbed and

reconstituted samples as described by Vaid et al. (1995a,1995b).

Liquefaction of sand is associated with the tendency forthe sand skeleton to contract under shearing. When suchcontraction is prevented or curtailed by the presence of alow-compressibility fluid such as water in the pores whichcannot escape, a large reduction in effective stress due to arise in pore pressure occurs, leading to a significant loss instrength and stiffness. This behaviour can occur under staticor cyclic loading conditions and is referred to as liquefac-tion. Liquefaction induced by static or monotonic loadingcan lead to flow slides in loose saturated sand slopes. Thesecan be triggered by small additional static loading or in-creases in groundwater levels. Many failures of slopes, par-ticularly those involved with mine waste disposal, haveoccurred in this manner. Cyclic liquefaction is commonlytriggered by earthquake loading and can lead either to a flowslide if the soil is loose, or to lateral spreading in denser

Can. Geotech. J. 37: 543562 (2000) 2000 NRC Canada

543

Received January 4, 1999. Accepted April 13, 2000.

P.M. Byrne.1 Department of Civil Engineering, TheUniversity of British Columbia, 2324 Main Mall, Vancouver,BC V6T 1Z4, Canada.1Corresponding author.

I:\cgj\Cgj37\Cgj03\T00-042.vpWednesday, June 28, 2000 3:16:46 PM

Color profile: Generic CMYK printer profileComposite Default screen

sands. Much of the damage in the San Francisco area thatoccurred during the Loma Prieta earthquake of 1989 wasdue to cyclic liquefaction, as was the great damage to theport facilities at Kobe during the 1995 earthquake.

Flow slides and large movements are controlled by thestressstrain relation and strength of the soil under un-drained loading. A major part of the CANLEX project in-volved the determination of these properties from in situ andlaboratory testing. These properties could then be used inanalyses to predict stability and deformations of soil struc-tures, allowing safe design of new structures and retrofit ofexisting structures.

The field event provided a test for numerical modellers topredict liquefaction movements under ideal conditions. Tohelp with the design of the field event and to help calibratethe numerical analyses, a number of centrifuge model testson sand embankment structures including one similar to thefield structure were carried out at the Centre for Cold OceanResources and Engineering (C-CORE) as described by Phil-lips and Byrne (1998). The challenge then was as follows:given the soil properties, could numerical modellers predictthe response of the centrifuge test and, more importantly,could they predict the response of the field test embank-ment?

Field event

The intent of the full-scale event was to statically trigger aliquefaction flow slide by rapidly loading a loose saturatedsand deposit. This involved the construction of a test em-bankment over the loose layer. The event is described in de-tail by Byrne et al. (1995a) and Robertson et al. (1996). Abrief description will be given here and is illustrated in Fig. 1a.

An abandoned borrow pit at the Syncrude Canada Ltd.site (J-pit) was used to carry out the field event. The founda-tion sand was placed hydraulically into standing water up toelevation 318 m and was referred to as beach below watersand (BBW sand). A level platform was then formed at ele-vation 321 m by placing tailings sand above the water andwas referred to as beach above water sand (BAW sand). Aclay dyke 8 m high with side slopes of 2.5:1 (horizontal tovertical) was constructed slowly over the tailings so as to al-low drainage of the sand to occur. A 10 m high compactedsand cell containment structure was then constructed to forman enclosure. Rapid loading was brought about by pumpingtailings (contained sand) behind the clay dyke.

The site was characterized as described by Hofmann et al.(1996a). This involved in situ testing (seismic cone penetra-tion testing, standard penetration testing with energy mea-surements, self-boring pressuremeter testing, and geophysicallogging), undisturbed sampling using in situ ground freez-ing, and laboratory testing of the samples. The targetsand layer was found to be very loose with an average(N1)60 3.5. Laboratory testing of undisturbed samplesshowed this very loose material to be generally strain soften-ing in both simple shear and extension loading, but generallystrain hardening in compression loading.

Five lines of instrumentation (Fig. 1b) were installed un-der the test embankment extending to about 30 m beyondthe toe of the clay dyke, as described by Hofmann et al.(1996b). Each instrumentation line contained piezometers,

tilt meters, piezo-settlement points, and surface-settlementpoints. In addition, a remote optical survey system was usedto monitor surface movements.

The field event began on 18 September 1995. Water andtailings were poured into the cell behind the clay embank-ment. The water level was raised to a height of 7.5 m andthe sand was placed to a height of 7 m. Loading took about36 h.

As a result of rapid loading, the clay dyke experienced amaximum displacement of 0.054 m at the toe, with an aver-age movement of 0.020 m (Natarajan et al. 1996). Hence,displacements due to loading were small. In terms of porepressures, the highest piezometric head (6.7 m) was mea-sured under the upstream slope of the clay embankment. Be-neath the crest of the dam the average piezometric head was3.5 m (Hofmann et al. 1996a). It is evident that these excesspore pressures were not sufficient to trigger liquefaction inthe foundation layer underneath the dyke and beyond its toe.

Centrifuge tests

A number of centrifuge tests were carried out to investi-gate the static flow liquefaction potential of embankmentssupported on very loose saturated sand layers. The purposeof these tests was (i) to serve as models to help design thefield event, and (ii) to provide a data base from which to cal-ibrate the numerical models. These tests are described in de-tail by Phillips and Byrne (1993, 1994). Test 1 (Phillips andByrne 1994) was chosen for detailed description and analy-sis because it resembled the proposed field event mostclosely.

A profile of the model test embankment is shown inFig. 2. During the initial loading stage, this structure wassubjected to an acceleration field of 50g. The accelerationfield was brought about in five increments of 10g each, al-lowing for pore-pressure dissipation between increments.Canola oil was chosen as the pore fluid to delay the rate ofpore-pressure dissipation. Delaying the rate of dissipationwas necessary to promote a near undrained type of responseduring the subsequent loading stage. Under an accelerationfield of 50g, the model structure corresponds to a 10 mdepth of foundation layer supporting a 5 m high embank-ment. After self-weight compression, the sand had a relativedensity of 29% (Phillips and Byrne 1994).

While in flight and under this acceleration field, a pres-sure load of 60 kPa, corresponding to about 3.5 m of soil,was applied on the crest of the slope, followed 2.6 s later byanother load of 60 kPa. Thus the loading was equivalent tothe rapid application of 7 m of soil over a 10 m target layer.The model was instrumented, with five pore-pressuretransducers, PPT1PPT5, and two displacement transducers,LDT18 and LDT19, as shown in Fig. 2.

Figure 3 shows the response of the centrifuge model toloading. Pore-pressure transducers PPT2 and PPT3, whichwere located directly beneath the load, registered sharpincreases in pore pressures followed by rapid drops undereach of the loading increments (Fig. 3a). However, a verydifferent pore-pressure response can be noted below the toeof the slope, as registered by PPT4. Here, the pore pressurerose rapidly but remained almost unchanged with applicationof the second load increment.

2000 NRC Canada

544 Can. Geotech. J. Vol. 37, 2000



2000

NRC

Canada

Byrne

etal.

545Fig. 1. (a) Plan and cross-sectional views of liquefaction event site. BAW, beach above water; BBW, beach below water. (b) Plan view of instrumentation lines through em-bankment.



2000 NRC Canada


Fig. 2. Initial centrifuge model and instrumentation configuration. All dimensions and vertical scale in mm. PPT, pore-pressure trans-ducer; LDT, longitudinal displacement transducer.

Fig. 3. Centrifuge model response to loading: (a) pore pressure during loading, (b) surface settlement during loading, (c) deformationfield due to first surcharge of 60 kPa.



The vertical movements recorded by LDT18 (on the crest)and LDT19 (in the toe area) are shown in Fig. 3b. LDT18indicated that the crest settled 4 mm under the first load andcontinued to deform upon application of the second load, upto 6.7 mm. Scaled to 50g, these settlements correspond to0.2 and 0.33 m, respectively. In contrast, LDT19 shows up-ward movement of about 3 mm, or 0.15 m scaled to 50g.

The pattern of displacements after the first load was ap-plied but prior to application of the second load is shown inFig. 3c. The pattern is deep seated and consistent with a liq-uefaction failure.

Liquefaction response of Syncrude sand

Tests carried out by Vaid et al. (1995a) showed that theundrained response of loose Syncrude sand to first-timeloading is strongly influenced by the direction of loadingrelative to the direction of deposition (vertical), i.e., the an-gle a

s

of the major principal stress, s 1, as shown in Fig. 4.Triaxial compression (a

s

= 0) as compared with triaxial ex-tension (a

s

= 90) is shown in Fig. 4a, which shows that re-sistance in compression is strain hardening, whereas inextension it is strongly strain softening. Simple shear load-ing, in which a

s

varies from 0 to about 45 during the test,is neither hardening nor softening, and its ratio of undrainedshear strength to vertical consolidation stress (su/s vc ) is ap-proximately 0.2. Hollow cylinder tests (Fig. 4c) show the re-sponse for a

s

values of 0, 30, 45, and 90 and illustrate theenormous effect of the direction of principal stress on un-drained shear response. For a

s

= 0, the undrained shearstrength ratio (su/po , in which po is the effective meannormal stress) is approximately equal to 0.3, whereas fora

s

= 90, su/po 0.06, i.e., a factor of five difference instrength. Since the direction of major principal stress willrange from extension to simple shear to compression in boththe model and field events, it is very important that this beconsidered in an analysis.

Numerical modelling

In stress deformation analyses of soil structures, equilib-rium and compatibility of every element in the domainshould be satisfied as well as the boundary conditions, andthese constraints must be fulfilled for the appropriate ele-ment stressstrain relations. Equilibrium, compatibility, and

the boundary conditions can be reasonably satisfied usingfinite element or finite difference formulations. Many com-puter codes exist for carrying out such calculations, e.g.,PISA (Chan 1997) and FLAC (Cundall 1995). The difficultyarises in modelling the complex stressstrain relations ofsoil, particularly under undrained conditions when liquefac-tion can occur, as was shown in Fig. 4. The commonly usedelasticplastic models in which the soil is assumed to be iso-tropic and elastic below the failure envelope and plastic atfailure will not simulate the strain softening observed inmonotonic liquefaction. Two basic approaches were taken tomodel the fundamental element liquefaction response.

The University of British Columbia (UBC) approachA fundamental plasticity approach that captures the be-

haviour of the sand skeleton including its plastic contractionupon shearing was developed at The University of BritishColumbia. The approach is described by Byrne et al.(1995b) and Puebla et al. (1996, 1997) and will be referredto here as the UBC approach. When the volume of the sandskeleton is constrained, either by the presence of water inthe pores that cannot escape or by constrained boundariessuch as in the simple shear test, plastic contraction uponshearing can lead to a large reduction in effective stress anda strain softening shear stress shear strain response. Theparameters for such a model can be obtained from drainedshear tests in which shear and volumetric strains are mea-sured or from undrained tests in which pore pressures aremeasured in place of volumetric strains. The model can beused to solve the drained or undrained response as well asthe coupled stressflow problem. In addition, the plasticcomponent of the model is anisotropic, so the marked differ-ence between the undrained response observed in triaxialcompression and that observed in extension can be captured.This was considered to be very important for both the centri-fuge and the field event. The model parameters (Table 1)were selected to give reasonable agreement with the testdata over a wide range of consolidation stresses and stresspaths. The predicted response for isotropically consolidated(po = 100 kPa) triaxial compression and extension tests andsimple shear tests (vertical consolidation stress, s vc =100 kPa) are compared with laboratory test data in Fig. 5and show good agreement. In this approach, strain softeningand an undrained collapse line are predicted for extensionloading rather than specified and result from the softer

2000 NRC Canada

Byrne et al. 547

Parameter Undrained triaxial and simple shear testsElastic bulk modulus number, kBe 300Elastic shear modulus number, kGe 300Elastic bulk modulus exponent, me 0.5Elastic shear modulus exponent, ne 0.5Plastic shear modulus number, kGp 310Plastic shear modulus exponent, np 0.67Peak friction angle, f f 33.7Constant volume friction angle, f cv 33.0Failure ratio, RF 0.95Factor of anisotropic plastic response, F 0.32

Table 1. Model parameters used in the UBC approach to simulate undrained behaviour of loose Syncrudesand.



plastic modulus used in extension. The FLAC computercode was used with this model.

University of Alberta (U of A) collapse-surfaceapproach

In this approach, the undrained strain softening response ofloose sand depicted in Fig. 6 is captured directly in the model.

The approach, referred to here as the U of A approach, was de-veloped at the University of Alberta by W.H. Gu and D.H.Chan and is described in more detail by Chan et al. (1995) andCathro and Gu (1995). The initial response is captured using anelastic formulation with conventional Skempton A and B un-drained parameters to generate pore pressures. Strain softeningis captured by specifying a collapse surface in terms of plastic

2000 NRC Canada


Fig. 4. Undrained monotonic behaviour of water-pluviated Syncrude sand: (a) isotropically consolidated, triaxial compression and ex-tension tests; (b) simple shear test (K0 consolidation); (c) isotropically consolidated, hollow-cylinder torsion test (modified from Vaid etal. 1995a). b = ( s 2 s 3)/( s 1 s 3); s vo , effective confining stress; ec, voids ratio at s vo ; po , effective mean normal stress; s 1, s 2 , ands 3, major, intermediate, and minor principle stresses, respectively; e 1 and e 3, major and minor principle strains, respectively;a

s

, angle between s 1 and the vertical.



parameters of cohesion and friction. By assuming that the re-sidual strength, su, is related to the consolidation pressure, po ,the collapse surface can be moved up or down to reflectchanges in void ratio arising from drainage. In addition, boththe residual strength and the slope of the collapse surface canbe varied with the direction of the major principal stress to cap-ture the response characteristics shown in Fig. 4, i.e., the col-lapse surface could also depend on the direction of loading aswell as the void ratio or effective consolidation pressure. Thestrain softening is captured by specifying strength as a functionof shear strain. The pore pressures (u) on the collapse surface

are computed by subtracting the predicted effective stresses(s ) from the total stresses (s ), i.e., u = s s .

Analysis of centrifuge tests

The University of British ColumbiaThe gradual buildup of stress during centrifuge swing

up was not modelled. The effective stresses due to selfweight under an acceleration field of 50g were computed byassuming a fully drained response and using the UBCstressstrain model described in detail by Byrne et al.

2000 NRC Canada

Byrne et al. 549

Fig. 5. UBC approach: element undrained response of reconstituted samples: (a) triaxial tests; (b) simple shear test. t , shear stress; u1,pore pressure.



(1995b) and Puebla et al. (1997). The densities of the sandwere based on the laboratory data reported by Phillips andByrne (1993). Two different densities of the sand were used:a submerged density of 0.96 t/m3 below the oil table, and atotal density of 1.87 t/m3 above the oil table.

The response to rapid loading was modelled as undrained.The undrained response of the foundation sand observed inlaboratory element tests over a range of stress states andpaths was captured with the stressstrain model using theparameters listed in Table 1. The slope was then modelled asa collection of such elements. The boundary conditions onthe model were as follows: zero horizontal displacements onthe vertical boundaries, and zero vertical displacements onthe bottom horizontal boundary, as shown in Fig. 7. The wa-ter table (oil table) was horizontal and coincided with thesurface of the target layer.

The rapid loading was simulated by a pressure of 60 kPa,increased shortly afterwards to 120 kPa, applied at the crestof the slope under undrained conditions. A very stiff andweightless structural beam member connected the loadednodes simulating the effect of the rigid weight. A compari-son between the measured and predicted pore pressures anddisplacements at the monitored points in the centrifuge testis presented in Table 2, which shows reasonable agreementbetween measured and predicted values.

The displacement pattern upon application of the firstweight is shown in Fig. 8. It compares reasonably well withthe pattern observed in the actual test (Fig. 3c) from the firstweight. Both patterns show that the region directly beneath

the loading plate was mainly sheared in compression mode,whereas in regions below the slope and beyond the toeshearing was mainly in simple shear and extension modes. Astrain-hardening type of response was predicted by the nu-merical model for elements making up the compressionregion, whereas a strain-softening type of response was pre-dicted for elements in the simple shear and extension re-gions.

University of AlbertaTwo separate analyses using the collapse-surface approach

were carried out, one by Chan et al. (1995), and one byCathro and Gu (1995). These analyses are described in moredetail by the authors.

2000 NRC Canada


Fig. 6. Simplified undrained model for liquefiable soils (after Cathro and Gu 1995). p , mean normal effective stress; q, deviatoricstress; D u, excess pore-water pressure. Model parameters: F ss, angle of steady state line; F pk, angle of peak strength line; F cs, angle ofcollapse surface; e p, equivalent plastic strain; SS, steady state strength on an undrained p q plane; K, hyperbolic strain softeningcurve; Kp and Kr , peak and residual strengths, respectively; a and b, hyperbola parameters.

Measured PredictedPore pressure (kPa)

PPT2 52 45PPT3 80 80PPT4 60 52

Longitudinal displacement (m)LDT18 (scaled to 50g) 0.34 0.39LDT19 (scaled to 50g) >0.14 0.34Note: PPT, pore-pressure transducer; LDT, longitudinal displacement

transducer (see Fig. 2).

Table 2. Comparison of measured and predicted (UBC approach)results in centrifuge test 1.



Chan et al. (1995) analysisThe centrifuge experiments were modelled using the finite

element program PISA. The program incorporates a col-lapse-surface plasticity model for modelling undrained de-formation of sand. A two-dimensional plane strain finiteelement mesh was used to model the prototype embankment.In performing the analysis, an initial stress field was incor-porated using the switch-on-gravity technique. This is simi-lar to the process of applying the centrifugal loading on thesoil. The application of the loading by the drop weight wasmodelled using a uniform pressure boundary. The numericalmodel assumes that the soil deforms in an undrained mannerwith pore-water pressures determined from the pore-pressureparameters. Stresses in the soil are determined and thencompared with the collapse surface, which is a function ofthe undrained shear strength of the material. If the stressesresult in a collapsible state, collapse analyses will be per-formed and the stresses and pore pressures will be redistrib-uted to obtain a new equilibrium state. If the zone of thecollapsible material is sufficiently large, overall failure mayoccur.

The material parameters used in the analysis are shown inTable 3. These parameters are determined based on triaxialtest results. The pore-pressure parameters were back-calculated from the equation of Henkel (1960) using ob-

served pore-pressure response during swing up in thecentrifuge experiments.

Since it is known that the undrained shear strength in-creases with an increase in mean effective stress, a constantsu /po ratio of 0.2 was used in the analysis to incorporate thevariable su. The numerical model indicated failure of theslope after the application of the first weight. The calculateddistorted mesh at failure is shown in Fig. 9 and the surfaceprofile compares reasonably well with the observed shapeshown in Fig. 3c.

Cathro and Gu (1995) analysisCathro and Gu (1995) extended the collapse-surface ap-

proach for liquefaction analysis to account for volumechange due to pore-pressure dissipation using the consolida-tion theory of Biot (1941). With this approach, drainage dur-ing swing up of the centrifuge model was simulated. Inaddition, the loading stage was modelled by considering un-drained conditions during application of the first 60 kPa sur-charge load. The pore pressures generated during thisprocess were then allowed to dissipate until the applicationof the second 60 kPa surcharge load, causing an additionalundrained rise in pore-water pressure. Pore pressures fromboth loadings were then allowed to dissipate. The predictedpore pressures at the locations of PPT2 and PPT3 agreed

2000 NRC Canada

Byrne et al. 551

Fig. 7. UBC approach: boundary conditions used for numerical simulation of centrifuge test.

Fig. 8. UBC approach: deformation pattern of centrifuge model due to first surcharge of 60 kPa.



remarkably well with the measured values (Fig. 10a). At thelocation of PPT4, the predicted rate of pore-pressure dissipa-tion was too fast compared with the measurements(Fig. 10b). In addition, a greater peak value of pore pressureunder the second load increment was predicted. The vectorplot of final displacements obtained in the numerical analy-sis is shown in Fig. 10c and shows good agreement with thetest results.

Analysis of field event

The three modelling groups had simulated the field eventprior to the event taking place. These results were presentedat a CANLEX meeting held at Fort McMurray on 7 and 8September 1995, 10 days before the event took place. Basedon the centrifuge modelling and the site investigation carriedout to that time, both U of A groups predicted that the testembankment would fail when loaded. The UBC group pre-dicted that it would not. The main reason for the differenceat that time was the incorporation of the stress-path oranisotropic effect on strength and stiffness in the UBCmodel, i.e., the effect shown in Fig. 4. The results presentedhere are based on papers dated after the event had occurred.

The University of British ColumbiaThe field event was modelled in the analysis by simulat-

ing the construction and loading conditions. It was assumedthat all materials behaved in a fully drained manner duringplacement, except the contained sand (Fig. 1), which was as-sumed to behave as a heavy fluid. The clay dyke and com-pacted cell sand were modelled as elastic perfectly plasticmaterials and their parameters were based on Duncan et al.(1980) and Byrne et al. (1987). These parameters are listedin Table 4. The parameters for the stressstrain model usedto simulate the target foundation sand response were ob-tained by back analysis of laboratory test data from Phillipsand Byrne (1994) and Vaid et al. (1995b) and are listed inTable 1.

The boundary conditions on the numerical model were asfollows: zero horizontal displacements on the verticalboundaries at x = 0 and 300 m (Fig. 1), and zero horizontal

and vertical displacements on the bottom boundary (naturalground surface, Fig. 1). The water table was horizontal andcoincided with the surface of the target layer (BBW sand)before the load was applied. After loading, the water levelwas assumed to coincide with the crest of the embankmenton the upstream side of the clay dyke and with the surface ofthe sand layer at elevation 321 m on the downstream side(Fig. 1a).

The rapid placement of the contained sand comprised theloading. Under this loading, two different assumptionswere made regarding the drainage conditions of the targetsand: (i) undrained, and (ii) coupled stressflow. This led totwo separate sets of analyses and results.

The predicted displacement pattern and contours of totalpore-water pressure obtained by numerical simulation underundrained conditions are presented in Fig. 11. The maxi-mum predicted horizontal displacement occurred in the toearea and had a magnitude of about 10 cm. The pattern ofpredicted displacements shown in Fig. 11a indicated thatsignificant displacements would occur to the full depth be-neath the upstream side of the dyke and would drop rapidlywith depth beneath the downstream slope of the embank-ment. Beyond the toe of the dyke, the predicted movementswere small. The contours of predicted total pore-water pres-sure presented in Fig. 11b show large pore pressures beneaththe applied load, but beyond the downstream slope of thedyke the effects of the load on pore pressure were not signif-icant. Lack of excess pore-water pressure beyond the toe ofthe dyke was the likely reason for the small displacements inthis region.

The coupled stressflow was simulated by allowing thegenerated pore pressures to dissipate during and after loadapplication. Once the contained sand (heavy fluid) was placedrapidly, its surface became a drainage boundary with porepressure equal to zero. The surface of the tailings down-stream of the toe was also specified as a drainage boundarywith pore pressure equal to zero. The vertical boundaries atx = 0 and 300 m (Fig. 1) and the bottom boundary (naturalground surface) were considered impermeable. Seepage flowinto all the zones within the boundaries was considered. Thehydraulic conductivity of the foundation sand was based on

2000 NRC Canada


Material B (kPa) G (kPa) c (kPa) f () kx (m/s) ky (m/s) r (t/m3)Clay dyke 30 000 30 000 30 0 1108 1108 2.0Sand cell 30 000 30 000 0 36 8106 5107 2.1Slurry 10 000 100 2 0 1104 1104 2.2Soft clay 10 000 2 000 10 0 11011 11011 1.6

Note: B, bulk modulus; G, shear modulus; c, cohesion; f , friction angle; kx, horizontal hydraulic conductivity; ky, vertical hydraulic conductivity; r ,density.

Table 4. Material parameters used in the UBC liquefaction analysis of the field event.

g b (kN/m3) E (kPa) n F ss () F cs () a Ao Am a su/po9.00 9000 0.3 33 24 0.001 0 5 0.3 0.2

Note: g b, submerged unit weight in Canola oil; E, Youngs modulus;

field experience at the Syncrude site which indicated that thehorizontal hydraulic conductivity (kx) was 15 times greaterthan the vertical hydraulic conductivity (ky). The valuesused for the analysis were kx = 7.5 106 m/s and ky =5.0 107 m/s. The hydraulic conductivity of the clay dykewas assumed to be k = 1.0 108 m/s (Cedergren 1989), andthe contained sand (heavy fluid) was considered to have ahydraulic conductivity k = 1.0 104 m/s (Cedergren 1989).

Results of the coupled stressflow analysis are shown inFig. 12. The maximum predicted horizontal displacementoccurred in the toe area and had a magnitude of about 5 cm.The pattern of displacements is presented in Fig. 12a and thecontours of predicted total pore-water pressure after thesteady state seepage condition was reached are shown inFig. 12b. The patterns of displacements and pore pressureare similar to those obtained for the undrained case; how-ever, the values are smaller. Allowing for dissipation of theexcess pore pressures leads to predicted displacements ofabout half the values predicted for undrained conditions.

The measured and predicted values of excess pore-waterpressure for both undrained and coupled stressflow analy-ses are compared in Table 5. The locations at which themeasurements were taken are shown in Fig. 13. In general,the predicted values of excess pore pressure are somewhathigher than the measured values.

The predicted and measured displacements in the toe areaare also shown in Table 5. The prediction from the coupledstressflow analysis agrees well with the measurements,whereas the undrained prediction was twice that of the mea-sured value.

University of AlbertaThe field event was modelled using the finite element

computer program PISA and the collapse-surface approachdescribed earlier. A certain depth of the foundation was in-cluded in the finite element mesh, and the mesh was ex-tended laterally far enough to minimize boundary effects.Initial in situ stresses in the tailings sands were generatedusing the switch-on-gravity technique. The material parame-ters used in the analysis are shown in Table 6. These param-eters were determined based on field and laboratory tests.Strength and deformation parameters for the tailings wereestimated from laboratory test results and the average coneresistance obtained during site characterization. Based on thesite-characterization information, the unit weight andstrength parameters for the top 3 m of the tailings were as-sumed to be slightly higher. Pore-pressure parameters used

in the undrained analysis are the same as the values used inthe U of A analyses of the centrifuge tests.

Modelling of embankment constructionThe construction of the clay embankment over the sub-

merged tailings was simulated in three ways representingdrained, undrained, and partially drained conditions. Adrained analysis was performed to simulate slow construc-tion of the embankment. This may not represent the realconditions due to the fact that a long time is required forpore pressures to completely dissipate. However, the drainedcondition represents one extreme case in the evaluation ofthe stability of the embankment. The finite element results ofthis analysis indicated that no significant deformationswould occur during embankment construction. An undrainedeffective stress analysis was carried out to model the rapidconstruction of the embankment over the tailings with nodissipation of excess pore-water pressures in any of the con-struction lifts. This represents the other extreme conditionthat may be experienced by the tailings sands. Results of theanalysis showed that failure would occur at an early stage ofembankment construction. Failure was indicated bynonconvergence of the numerical solutions.

The above two cases represent extreme conditions thatmay be induced in the sand during embankment construc-tion. Realistically, partial drainage (or complete drainage,depending on the permeability of the tailings sand and therate of construction) will occur. To simulate partial dissipa-tion of pore pressures between each lift of fill placement, anundrained analysis was carried out with excess pore pres-sures in the sand being reduced between lifts. The pore pres-sure was reduced to sufficiently low values that would allowstable construction of the embankment. Piezometric headmeasurements taken during embankment construction con-firmed that dissipation of pore pressures in the foundationsand did occur between lifts. Results of the partially drainedanalysis are plotted in Fig. 14, which represents the maxi-mum allowable excess pore-water pressures at the mid-depthof the tailings sands (under the centreline of the embank-ment) to avoid instability during the embankment construc-tion. This figure was provided to Syncrude Canada Ltd. as aguide for construction control during the embankment con-struction. During construction, except for two piezometers inline 5, the responses of each piezometer located in the tail-ing sands did not exceed the recommended values given inFig. 14. This resulted in safe construction of the clay em-bankment.

2000 NRC Canada

Byrne et al. 553

Fig. 9. U of A approach (Chan et al. 1995): pre- and post-failure profiles of centrifuge test.



Modelling of loading due to contained tailingsA number of analyses were carried out to investigate the

effect of contained tailings on the stability of the embank-ment. In these analyses, contained tailings were assumed tobe placed after all excess pore-water pressures in the tailingssands due to the embankment construction had dissipated.This assumption was made based on the observed responseof the piezometers, which indicated relatively fast dissipa-tion of the excess pore water pressure in the tailings sands.The analyses are categorized into three main cases:

Case 1: The contained tailings were assumed to be placedrapidly into the area between the embankment and the com-pacted sand dyke so that the tailings imposed undrainedloadings on the structure. No dissipation of excess pore-water pressures was assumed during filling.

Case 2: Steady state seepage in the tailings sands was as-sumed during filling of the contained tailings. In this case,

the tailings sands and the embankment were loaded underdrained conditions.

Case 3: The contained tailings were assumed to be placedin three lifts, with reduction of excess pore pressures in thetailings sands between lifts. This analysis was performed tostudy the effect of the loading rate on the stability of the em-bankment.

In situ testing, including cone penetration tests (CPT) andstandard penetration tests (SPT), before the embankmentconstruction indicated that the tailings sands were generallyloose, although highly nonhomogeneous. Based on the insitu and laboratory test results, the average undrained steadystate strength ratio of the tailings sands (su /po ) before theembankment construction was estimated to be about 0.1. Ra-tios of su /po in the range of 0.10.2 were used in the analy-sis to examine embankment sensitivity to this parameter. Toinvestigate the effect of higher strengths in the top 3 m of

2000 NRC Canada


Fig. 10. U of A approach (Cathro and Gu 1995): computed response of centrifuge model to loading. (a) Pore-pressure response atPPT2 and PPT3. (b) Pore-pressure response at PPT4. FEM, finite element method.



the tailings sand, analyses were carried out with higher su /poratios for the top 3 m of the tailings. In addition, an increasein the su /po ratio of the tailings sands under the clay em-bankment was expected due to consolidation. In this regard,the undrained strength, su, beneath the embankment wouldincrease under constant su /po because p increases as the em-bankment is placed and dissipation occurs. It was felt thatadditional densification resulting in an increase in su /powould occur due to vibration associated with compaction ofthe overlying dyke. To examine this effect for the tailingsunder the embankment, cases 1, 2, and 3 were analyzed us-ing different su /po ratios, as summarized in Table 7.

Results of analyses by the University of AlbertaBased on the information summarized in Table 7, the ma-

jor findings of the liquefaction analyses were as follows:(1) Rapid filling of the contained tailings behind the clay

embankment, with no dissipation of excess pore-water pres-sures during filling, would trigger liquefaction flow failurein the tailings sands if the average undrained steady statestrength ratio of the tailings sands did not exceed 0.1.

(2) Liquefaction flow failure (if any) would occur as a re-sult of undrained loading before steady state seepage is es-tablished in the tailings sands.

(3) Results from the case 3 analysis indicated that if load-ing was not fast enough, a liquefaction failure would not oc-cur. Rapid loading of the contained tailings is essential inorder to trigger a flow failure. Therefore it was recom-

mended that the contained tailings should be filled continu-ously and be completed in about 24 h to minimize dissipa-tion of excess pore-water pressures in the tailings sands.Figure 15 shows contours of computed excess pore-waterpressures at the end of loading for the case 3 analysis.

Although the loading rate was relatively fast from a feasi-bility point of view, it was not fast enough to induce highenough excess pore pressures in the tailings to reach a stateof instability. It is important that the rate of loading be stud-ied together with the rate of dissipation of excess pore-waterpressure. Piezometer measurements during construction ofthe clay embankment showed that the tailings sands arehighly permeable. Almost half of the induced pore-waterpressures were dissipated after 12 h from the beginning oflift construction, and piezometer response during loading be-hind the embankment indicated that a large amount of ex-cess pore pressure had been dissipated during loading.Results of the case 3 analysis showed that if the loading wasnot fast enough (i.e., most of the excess pore pressure wasallowed to dissipate during loading), no liquefaction flowfailure would occur. Table 8 compares the observed and cal-culated changes in water level for case 3 in line 1 of the in-strumented section; the piezometer locations in line 1 areshown in Fig. 16. The instrumented section along line 1showed the highest pore-pressure response during the exper-iment.

Figure 17 compares the measured excess pore-water pres-sure at piezometer P18A during loading with the calculated

2000 NRC Canada

Byrne et al. 555

Fig. 10 (concluded). (c) Displacement field from finite element analysis.



excess pore-water pressure for cases 1, 2, and 3. As indi-cated, the computed excess pore-water pressure from case 1,which assumes fully undrained conditions, is much higherthan the observed excess pore-water pressure. However, theexcess pore-water pressures calculated from cases 2 and 3were close to the measured values. These results indicatethat the assumption of fully undrained loading was not real-istic due to the relatively high permeability of the tailingssands.

(4) If the su /po ratio of the tailings sands under the em-bankment was increased to 0.2 due to construction of theclay embankment, failure would not occur.

(5) Figure 4 showed the strong dependency of strength onstress path. Based on these results, it is useful to investigatestress paths in the tailings sands during loading. Figure 18shows contours of the angle (a

s

) between the major princi-pal stress (s 1) and the depositional (vertical) direction. Asindicated, most of the tailings, including the tailings underthe upstream side of the embankment, were sheared predom-inantly in compression. However, towards the downstreamside, some regions were subjected to simple shear type load-ing and only the surface material near the toe was subjectedto the extension mode of shearing. Therefore, it is reason-able to conclude that a rigorous liquefaction analysis basedon the above undrained steady state strength ratios (0.3 for

compression, and 0.06 for extension) most likely will resultin nonfailure of the embankment.

Discussion of field event predictions

Two very different techniques were used by the modellers:(1) The UBC approach is a fully coupled plasticity proce-

dure that can capture the sand skeleton behaviour over awide range of loading paths and drainage conditions, i.e.,drained, undrained, and coupled stressflow conditions. Themodel parameters are selected to give the best fit to labora-tory element test data.

(2) The U of A approach is a procedure in which the un-drained strength and slope of the collapse line are specified,together with a strain-softening function and a stress-redistribution procedure to assure that elements do not vio-late the collapse line and the steady state strength. Theundrained steady state strength can be increased to accountfor consolidation or loading path, and the slope of the col-lapse line can be adjusted with the direction of loading.

Despite the differences in analysis procedures, the resultswere found to depend more on the assumption made aboutthe stressstrain and strength of the soil than on the proce-dure itself.

2000 NRC Canada


Fig. 11. UBC approach: CANLEX embankment predicted response under undrained loading. (a) Displacement pattern. (b) Contours oftotal pore-water pressure.



Both penetration testing and results of undisturbed sam-pling and testing were available at this site. The laboratorytesting showed that direction of loading was a key factoraffecting the shear resistance of the target sand. If loaded incompression (direction of deposition), the sand was predom-inately strain hardening and had an average residual strengthratio su /po 0.3. When loaded in extension, the sand waspredominately strain softening and had an average residualstrength ratio su /po 0.06. The key question is what valueof su /po should be used in the modelling.

The UBC approach amounted to using a variable su /pobased on the laboratory test data, so zones of compressionloading had su /po = 0.3 and zones of extension loading inthe toe area had su /po = 0.06. However, because the strengthin the compression zone upstream of the crest of the dykewas so high, the stresses in the potential strain-softeningzone in the toe region never rose to their peak value and sodid not strain soften. Hence, high pore pressures in this zonewhich could lead to a liquefaction failure did not develop.

The U of A modellers based their su /po value on penetra-tion-resistance values and the laboratory data, and whilethey did use a variation in su /po to reflect looser and denserzones, they did not include the effect of the direction ofloading. They concluded that, had they considered this, theywould have predicted that the dyke would not have failed,even if the loading had been undrained.

Although the analyses carried out were quite different inprinciple, the key item affecting stability was the assumptionconcerning the residual strength. If su /po = 0.1 was se-lected, which, based on current practice, would not be un-duly conservative for such a loose sand, an undrained failurewould have been predicted by both modelling groups. Thefield test showed that advantage can be taken of the largevariation in residual strength with direction of loading whendesigning against a static liquefaction failure.

2000 NRC Canada

Byrne et al. 557

Fig. 12. UBC approach: CANLEX embankment predicted coupled stressflow response. (a) Displacement pattern. (b) Contours of totalpore-water pressure.

MeasuredCoupled stressflow prediction

Undrainedprediction

Excess porepressure (kPa)P14A 31.6 37.6 40.3P10A 11.8 21.3 25.7P06A 8.0 7.2 7.3

Displacement (cm)Toe of dyke 5 5 10

Table 5. Measured and predicted (UBC approach) excess porepressures and displacements for the Syncrude site, line 2.



2000 NRC Canada


Fig. 13. Piezometer locations in liquefaction event site along line 2 instrumented section.

Fig. 14. U of A approach: maximum excess pore pressure at mid-depth of tailings sands (under centreline of embankment) for safeconstruction of dyke.

Material g (kN/m3) E (kN/m2) n F ss () F cs () Ao Am su /poTailings (lower 8 m) 8.95 14 000 0.49 34 24 0 5 0.2, 0.15, 0.1Tailings (top 3 m) 9.4 18 000 0.49 36 25 0 5 0.2, 0.15, 0.1

Note: (N, effective unit weight.

Table 6. Parameters of tailings sands used in the U of A analysis of the event.



2000 NRC Canada

Byrne et al. 559

su/po Case 1 Case 2 Case 3Constant su/po 0.1 Failure No failure No failure

>0.1 No failure No failure No Failuresu/po of the top 3 m of tailings is higher* 0.15 Failure No failure

>0.2 No failure No failure su/po of tailings right under embankment is higher 0.15 Failure No failure

0.2 No failure No failure *A value of 0.1 is used for su /po of the lower 8 m of tailings.A value of 0.1 is used for su /po of the tailings which are not located under the clay embankment.

Table 7. Summary of the analyses of loading (U of A approach) due to contained tailings.

Fig. 15. U of A approach: contours of excess pore pressures (kPa) at the end of loading in the case 3 analysis.

P18A P13A P13B PF19C2 PO9A PF19T2Observed water level (m) 6.8 3.4 3.8 1.8 1.6 0.7Calculated water level (m) 7.1 2.9 3.4 1.2 1.0 0.9

Note: See Fig. 16 for location of piezometers.

Table 8. Comparison between the observed and calculated (U of A) changes in water level for case 3 in line 1 of theinstrumented section.

Fig. 16. Location of piezometer tips of the instrumented section along line 1. WT, water table.



Summary

The knowledge of stratigraphy and material propertieswere very well known at the Syncrude test site. In particular,(i) the investigation and testing of the target sand were ex-tensive and much greater than is generally available in prac-tice, and direct measures of stressstrain and strength wereavailable from testing on undisturbed samples and indirectmeasures in the form of penetration resistance tests; (ii) thesite investigation, laboratory testing, and field inspection allindicated that the target tailings sands were generally in avery loose state prior to the field loading event; and(iii) model centrifuge data using the same Syncrude sand

were available. In light of this, how well did the modellersdo?

The results of the analyses showed that the key item con-trolling the stability of the test embankment was the residualstrength and its variation with direction of loading and drain-age conditions. The UBC modelling group incorporated thedirection of loading directly into their analyses and predictedresults both before and after the event in good accord withfield observations. They concluded that the dyke would notfail under undrained conditions and drainage effects wouldreduce movements by about 50%.

The U of A modellers initially based their residualstrength values on penetration tests and state of practice de-

2000 NRC Canada


Fig. 17. U of A approach: comparison of observed and computed pore pressures at piezometer P18A, line 1.

Fig. 18. U of A approach: contours of the angle (a ) between major principal stress ( s 1) and depositional (vertical) direction (a is indegrees).



sign, and concluded the test embankment would fail. Basedon the direct assessment of su from laboratory tests and tak-ing loading path into consideration, they later concluded thatthe test embankment should not fail under undrained condi-tions. They also concluded that drainage was significant dur-ing the event and would have a stabilizing effect.

In practice, engineers are required to give assurance that aslope will not fail or deform excessively. This is generallyachieved using a limit equilibrium analysis approach, appro-priate soil shear strengths, and an appropriate factor ofsafety. This simple approach is not really applicable to theliquefaction problem described here because the soil hasboth peak and residual strengths, and the level of strain re-quired to achieve these strengths must be considered. In ad-dition, the strength depends very markedly on the directionof loading. Consequently, a higher level of analysis, whichconsiders strain compatibility and equilibrium, is required toconsider these factors. This was the approach taken by bothmodelling groups. The main difference was in the steadystate strength used. Both modelling groups used a normal-ized undrained strength ratio (su /po ) approach; the main dif-ference lay in whether the effect of direction of loading onsu /po was considered or not. The current practice is not toconsider direction of loading effects and to use a residualstrength based on penetration resistance and field experi-ence, i.e., an indirect measure of strength. The analyses andfield event presented here show that this approach may beunduly conservative because (i) direction of loading is im-portant, (ii) the undrained strength in the compression zonemay be very much higher than assumed, and (iii) drainageleading to increased strength may be significant. All of thesefactors were present at the Syncrude test site and preventedthe occurrence of a failure where the current-practice ap-proach would have predicted a failure.

Generally, the problem is to assure, with some degree ofcertainty, that a failure will not occur. In doing this, low val-ues of shear strength based on penetration testing are usuallyused in analyses. On occasion, a direct measure of residualstrength based on recovery and testing of undisturbed sam-ples is undertake, e.g., Duncan Dam (Byrne et al. 1994). Atthe Syncrude site, the problem was to predict the likelihoodthat failure would occur. In such a case, a high estimate ofshear strength should be used in analyses, together with afactor of safety less than unity, to assure that failure willoccur. Although the analyses that took stress-path effectsand drainage into consideration indicated that a flow failurewould not occur, no reasonable engineer would have advisedthat the dyke would not fail and that safety precautions didnot need to be considered during filling of the test embank-ment.

The results of the field event and analyses indicate thatsteep slopes can be built over very loose sand, provided therate of loading is such that consolidation can occur. Oncebuilt, the resistance to shock loading is greatly influenced bythe direction of loading. In areas of vertical compression loading,the undrained strength ratio may be high (i.e., su /po > 0.3),even for very loose sand, whereas in regions of vertical extension(toe area) the strength ratio may be very low (i.e., su /po

canlex 4 modelling cgj

Documents