29

A particulate-scale investigation of cemented sand behavior

Y.-H. Wang and S.-C. Leung

Abstract: In this paper, triaxial tests and numerical simulations using the discrete element method (DEM) are combined to explore the underlying mechanisms of the unique behavior of artificially cemented sands. The experimental results show that strength enhancement, volumetric dilation, and the shear banding associated failure mode are observed in Portland cement sand; these features become more pronounced with increasing cement content. Different responses are found in gypsum-cemented sand even though both types of cemented sand specimens were prepared under very similar void ratios before shearing. The DEM simulations on the Portland cement sand were carried out under two particular arrangements (i.e., the use of small cementing particles and flexible membrane boundaries). The simulation results reveal that particles in the bonding network jointly share the loading and many micro force-chains associated with cementation are created. Compared with uncemented sand, a more stable and stronger forcechain complex subjected to smaller force concentration is formed in cemented sand, which gives rise to higher strength. Intensive bond breakage, concentrated relative particle movement, column-like force chains, great particle rotation, and high local porosity are found inside the shear band. The bonded clusters remain at large strains to help stabilize the particle arch and therefore to maintain the volumetric dilation. Key words: discrete element method, forcechain distribution, shear banding, volumetric dilation, bonded cluster. Resume : Dans cet article, des essais triaxiaux et des simulations numeriques utilisant la methode des elements discrets (DEM) sont combines pour explorer les mecanismes sous-jacents du comportement unique dans les sables cimentes artifi ciellement. Les resultats experimentaux montrent quon observe une bonification de la resistance, une dilatation volume trique, et un mode de rupture associes aux bandes de cisaillement dans le sable-ciment Portland; ces caracteristiques deviennent plus prononcees avec laugmentation de ciment. On trouve des reponses differentes dans le sable cimente au ` gypse meme si les deux specimens ont ete prepares avec des indices des vides tres similaires avant le cisaillement. Les si ` mulations en DEM sur le sable-ciment Portland ont ete realisees dans deux arrangements particuliers (c.-a-d., lutilisation ` ` de petites particules de cimentation et des frontieres de membranes flexibles). Les resultats de la simulation revelent que les particules dans le reseau de bandes partagent conjointement la charge, et plusieurs chanettes de microforces associees ` a la cimentation sont creees. En comparaison avec le sable non cimente, un complexe plus fort et plus stable de chanettes ` ` de forces assujetti a une concentration de forces plus petites se forme dans le sable cimente, ce qui donne naissance a une ` resistance plus elevee. A linterieur des bandes de cisaillement, on trouve des bris intenses de liens, un mouvement relatif des particules, des chanettes de forces sous forme de colonnes, de fortes rotations de particules, et une forte porosite lo ` ` cale. Les agglomerats lies survivent a de grandes deformations pour aider a stabilizer larche de particules et ainsi mainte nir la dilatation volumetrique. Mots-cles : methode delements discrets, distribution de chanettes de forces, bandes de cisaillement, dilatation volume trique, agregat lie. [Traduit par la Redaction]

Introduction
Cemented sands are widely found in nature, for example in aged sedimentary deposits. As a matter of fact, cementation is often observed during the early diagenesis process (Pettijohn et al. 1987). Natural cementation originates from different sources, for example, from the precipitation of calcite, silica, iron oxides, and even clays (Santamarina et al.
Received 13 December 2006. Accepted 8 August 2007. Published on the NRC Research Press Web site at cgj.nrc.ca on 8 February 2008. Y.-H. Wang1 and S.-C. Leung. Department of Civil Engineering, The Hong Kong University of Science and Technology, Hong Kong, China.
1Corresponding

author (e-mail: ceyhwang@ust.hk).

2001; Mitchell and Soga 2005). Apart from natural cemented soils, artificially produced cemented soils are often encountered in projects where soil is improved by mixing it with cement or chemicals. Although the formation of cemented soils is complicated, general features regarding the cementation effects on soil properties can still be observed in a collective way based on the experimental findings: (i) the strength (dynamic and static) and small-strain stiffness are enhanced (Dupas and Pecker 1979; Acar and El-Tahir 1986; Clough et al. 1981 and 1989; Saxena et al. 1988; Huang and Airey 1998); (ii) the stressstrain and volumetric response become relatively brittle and more dilative, respectively (Clough et al. 1981; Lade and Overton 1989; Abdulla and Kiousis 1997; Schnaid et al. 2001); and (iii) quasipreconsolidation pressure or yield stress can be observed in response to loading (Leroueil and Vaughan 1990; Airey
#

Can. Geotech. J. 45: 2944 (2008)

doi:10.1139/T07-070

2008 NRC Canada

30

Can. Geotech. J. Vol. 45, 2008 Fig. 1. The grain-size distribution of Ottawa 2030 sand.

1993). In addition, the mechanical responses of cemented soils are found to depend on the amount and nature of the cementing agents (Ismail 2000; Ismail et al. 2002). However, in these studies, only the macroscale responses are emphasized, and the associated underlying mechanisms often remain hypothesized, vague, or even undescribed. The main objective of this study is to reveal these puzzles in explaining cemented sand behavior with a focus on strength enhancement and volumetric dilation due to cementation. Experimental characterizations by drained triaxial compression tests and numerical simulations by the discrete element method (DEM) will be used in parallel to achieve this purpose.

Experimental details
Artificially cemented sand samples Obtaining high-quality samples of natural cemented sands from the field is difficult because cementation can sometimes be destroyed during sampling. In addition, identical samples of natural cemented sand cannot be replicated for parametric studies. Hence, artificially cemented sands are used in this study. The properties of the cementing agents are essential in determining the behavior of cemented sands, therefore two very different cementing agents were purposely selected (i.e., Portland cement and gypsum). One represents strong and stiff cementation whereas the other represents weak and soft cementing bonds. Ottawa 2030 sand is chosen as the sand matrix simply because its grains are uniformly sized and have a round shape. This characteristic can be easily modeled using DEM in numerical simulations. The grain-size distribution is shown in Fig. 1. The maximum and minimum void ratios were determined according to the BS 13774 (1990) standard. They are 0.749 and 0.481, respectively. Sample preparations and testing procedures Natural cemented sands exhibit a wide range of densities; however, to eliminate the density effects on the mechanical responses, the samples used in this study were purposely prepared in a loose state. All the samples also had a similar initial void ratio, e & 0.72, or a relative density of ~11% before shearing in the drained triaxial compression tests. The artificially cemented sands were produced by mixing cement slurry with Ottawa sand particles. The sand particles were thoroughly washed and oven dried at 100 8C for 24 h before use. The slurry was prepared by mixing water with Portland cement or gypsum powder in a ratio of 1:1 by weight. The mixture of cement slurry and sand particles was then put into a cylindrical sample mold. A specimen with a height of 140 mm and a diameter of 70 mm was prepared by 10 layer compaction following the method suggested by Ladd (1978) to ensure homogeneity. The compacted specimens were sealed into a plastic bag under a constant temperature of ~20 8C to minimize moisture loss during the curing process and cracking caused by temperature variations. The curing times for the Portland cement and gypsum-cemented specimens were 7 and 3 d, respectively. The cement content was determined by the ratio of the cement weight to the dry weight of the sand particles. The drained triaxial compression tests were performed using a CKC triaxial system (Li et al. 1988). Lubricated end

platens were used to reduce the friction between the end platen and the specimen. Carbon dioxide gas was circulated through the specimens for 45 min first to facilitate the saturation. However, this process was not used for the Portland cement specimens because carbon dioxide was expected to react with Portland cement through an acidalkali reaction, which could have affected the bonding strength. Deaired water was then circulated to saturate the specimen through a small pressure difference (5~10 kPa). Back-pressure saturation (200 kPa) was implemented in the last stage to ensure that the B-value was greater than 0.95. The designated confining pressure was applied to the specimen, and shearing was carried out with an axial-strain rate of 6%/h. To make the cementation effects more pronounced and to prevent severe bond breakage, a low confining pressure was used. The maximum values of 100 and 50 kPa were selected for the Portland cement and the gypsum-cemented samples, respectively.

Experimental results
Strength and failure mode Figures 2 and 3 demonstrate the stressstrain response of the Portland cement samples under the drained triaxial compression tests with different cement contents (1%, 2%, and 3%) and confining pressures (50 and 100 kPa). For comparisons, the responses of the uncemented sample are also presented. In the cemented samples, the peak strength and the associated strain-softening response become more distinct as the cement content increases or the confining pressure decreases. The uncemented samples, however, always follow a strain-hardening process. Adding cementation can markedly increase the strength and alter the stressstrain behavior. These findings are in accordance with previously published results, for example, Clough et al. (1981), Abdulla and Kiousis (1997), Schnaid et al. (2001), and Ismail et al. (2002), although the void ratio (before shearing) is not a control factor in these earlier tests. As shown in the images captured during testing, shear banding occurs with the strain-softening response (see picture B in Fig. 2). Also, the appearance of shear banding becomes more visible with increasing cement content, and it cannot be observed in the
#

2008 NRC Canada

Wang and Leung

31

Fig. 2. Stressstrain relationships of Portland cement sand samples under 50 kPa confining pressure. Note that the deformation characteristics are also demonstrated in the images captured during testing.

Fig. 3. Stressstrain relationships of Portland cement sand samples under 100 kPa confining pressure.

uncemented sample, which exhibits only a bulging type of failure (see picture D in Fig. 2). As demonstrated in Fig. 4, the influence on the stress strain behavior due to gypsum cementation is less when compared with the Portland cement samples under the same confining pressure of 50 kPa. This may be because gypsum bonding is naturally weaker and some cementation is destroyed during the stage of isotropic consolidation. The pronounced strength and stiffness enhancement can be found only under a lower confining pressure of 25 kPa (Fig. 5). All the gypsum-cemented samples are prone to exhibit a strain-hardening response or, more precisely, the difference between the peak and ultimate strength is very small, and strain localization is not observed (see Fig. 4, picture E). Volumetric response Figures 6 and 7 present the volumetric responses of the Portland cement samples. All the Portland cement samples

exhibit volumetric dilation upon shearing even with 1% cement content, whereas the uncemented sample shows volumetric contraction. As noted, all of the samples are prepared with very similar void ratios and are in a loose state. Therefore, the volumetric dilation is completely due to the cementation effect and not to the density effect. This dilative feature has been hypothetically attributed to cemented particles forming highly interlocked clusters (Saxena et al. 1988; Lade and Overton 1989); however, such a general explanation requires further elaboration (to be discussed later). The dilative response, as expected, is suppressed by confinement. Figures 8 and 9 present the volumetric responses of the gypsum-cemented samples; they are completely different from the response of the Portland cement samples. The volumetric contraction is higher than the amount measured in the uncemented specimens. This extra contraction increases with the gypsum cement content and may be attributed to crushing of the gypsum cementation. Gypsum cement is light (the specific gravity, Gs, is 2.3). A 3% or 5% cement content, which is determined by weight, implies that an appreciable volume of the sample is occupied by the gypsum cementing agent. To maintain a similar initial void ratio of the sample (i.e., e & 0.72), the sand particles themselves have to form an even looser packing compared with the uncemented sample. If the cementing agent of gypsum is easily crushable, greater contraction should result. Similar comments on the features of crushable gypsum cementation can also be found in Ismail (2000).

Details of the numerical simulations

Experimental results showed that cementation by Portland cement has a more pronounced influence on soil properties (i.e., increasing strength and changing the volumetric response from contraction to dilation). Numerical simulations
#

2008 NRC Canada

32

Can. Geotech. J. Vol. 45, 2008

Fig. 4. Stressstrain relationships of gypsum-cemented sand samples under 50 kPa confining pressure. Note that the deformation characteristics are also demonstrated in the images captured during testing.

Fig. 5. Stressstrain relationships of gypsum-cemented sand samples under 25 kPa confining pressure.

Fig. 7. Void-ratio variations of Portland cement sand samples in response to shearing (100 kPa confining pressure).

Fig. 6. Void-ratio variations of Portland cement sand samples in response to shearing (50 kPa confining pressure).

Fig. 8. Void-ratio variations of gypsum cemented sand samples in response to shearing (25 kPa confining pressure).

2008 NRC Canada

Wang and Leung Fig. 9. Void-ratio variations of gypsum cemented sand samples in response to shearing (50 kPa confining pressure). Table 1. Parameters of the three kinds of particles used in the DEM simulations. For soil particles Soil particle density (kg/m3) Initial porosity Radii of particles (m) Interparticle friction angle Particle/cap friction angle Normal contact stiffness (N/m) Shear contact stiffness (N/m) Contact stiffness between soil and membrane particles (N/m) For membrane particles Radius of membrane particle (m) Particle density (kg/m3) Normal bond strength (Pa) Shear bond strength (Pa) Normal contact stiffness (N/m) Shear contact stiffness (N/m)

33

2650 0.20 0.375103 1.475103 0.55 0.00 5107 5107 5106

were carried out to understand the associated underlying mechanisms of this behavior. The commercial DEM software, Particle Flow Code 2-Dimensional, PFC2D (Itasca Consulting Group Inc., Minneapolis, Minn. ) was used. The input parameters, such as the stiffness and bond strength in the normal and shear directions (i.e., the contact model), particle size and density, and friction between particles, are intimately coupled with the constitutive relations that govern the overall responses of cemented sand. Table 1 summarizes the parameters of the three kinds of particles used in the simulations (i.e., sand particles, cementing particles, and membrane particles). By combining these three kinds of particles, the numerical specimen can be formed, as illustrated in Fig. 10. The size of the numerical specimen is 140 mm in height and 70 mm in width. The simulations of each kind of particle are described in detail in the following sections. Simulations of sand particles The sizes of sand particles are generated according to the grain-size distribution of Ottawa 2030 sand. For the purpose of reducing the calculation time, the radii were scaled up 1.25 times to reduce the total particle numbers used in the simulations, which were about 2000 in total. The normal stiffness was set on the order of 1 107, which was inferred from the forcedisplacement relationship of quartz sand particles (Nakata et al. 1999) and from the relationship proposed by Potyondy and Cundall (2004). According to Bass (1995), the ratio between the bulk modulus and the shear modulus of quartzose material is close to 1, and, therefore, the same normal and shear stiffness were selected for the sand particles in the simulations. The input parameters were adjusted by comparing the simulation results with the measured responses of the intended physical specimen until similar behavior could be reproduced as suggested in Hazzard et al. (2000). The same method was also implemented in finding suitable input parameters for cementing and membrane particles. Simulations of cementing particles and associated bonding The interactions between the cementing agents and the soil grains are essential in reproducing true soil behavior in the simulations. A special arrangement between the cement-

0.5104 1800 110300 110300 5106 5106

For cementing particles (Portland cement) Particle density (kg/m3) 3150 Radius of cementing particle (m) 6.66105 Bond radius 1 Normal contact stiffness (N/m) 5107 Shear contact stiffness (N/m) 5107 Normal bond strength (Pa) 3106 Shear bond strength (Pa) 3106

ing particles and the sand grains was adopted, and it is shown in detail in Fig. 10a. The cementing particles are modeled with tiny particles deposited around the contacts of the large soil grains. This arrangement is similar to the real formation process of cementation as shown in Fig. 11: a strong capillary suction keeps pulling the cementing agents towards the particle contacts during the drying process until the cementation precipitates or crystallizes around the contacts. The interaction between sand grains and cementing particles is simulated better, and the amount of cement contents is controlled better with this arrangement. In the experiments, the cement content was quantified by the weight ratio. The same practice was also used in the simulations. The weight ratio was converted to a volume ratio or area ratio by relating it to the specific gravity (Gs) of the cementing agent. This area ratio was then used to regulate the number of cementing particles in the simulations for different cement-content specimens. For instance, the 3% Portland cement specimens require ~6000 cementing particles. The specific gravity used in the calculations is 2.65 for quartz (i.e., for the sand particles) and 3.15 for Portland cement. As indicated in Fig. 10b, the parallel-bond model was adopted in the simulations to regulate the contact model between the cementing particles and the sand grains. The parallel-bond model can be envisaged as a set of constant normal and shear stiffness uniformly distributed over a contact area (details are given in the users manual for PFC2D). The model is characterized by five parameters: normal and shear strength, linear normal and shear stiffness, and the
#

2008 NRC Canada

34

Can. Geotech. J. Vol. 45, 2008

Fig. 10. Characteristics of the numerical specimen: (a) special arrangements of the cementing particles; (b) parallel bonds between cementing particles and sand grains; (c) specimen used in the biaxial tests; (d) membrane particles.

Fig. 11. Features of Portland cement sand samples: (a) physical specimen used in the triaxial test; (b) loosely cemented sand particles; (c) cementation linking the sand particles together; (d) cementation formed at particle contacts. The sand image was taken through a microscope (Olympus SZH10 Research Stereo Microscope, Mellville, N.Y.).

bond radius as given in Table 1. The cementation bond was considered broken when the normal force or shear force exceeded the maximum resistance strength. Simulations of the membrane particles and applying a confining pressure Kuhn (1995), Corriveau et al. (1997), and Iwashita and Oda (1998) adopted the membrane particles in their DEM

simulations and suggested that using a flexible membrane boundary can capture the deformation characteristics particularly well in the development of shear banding. A similar simulation method is used here but further modified as illustrated in Fig. 10d. The flexible membrane is modeled by a string of small same-sized particles. These particles are linked by a strong but flexible contact bond so that the moment cannot be transmitted. The membrane-particle string
#

2008 NRC Canada

Wang and Leung

35

Fig. 12. Confining pressure applied to the flexible membrane boundary: (a) the equivalent force applied on the membrane segment; (b) the force shared by the individual membrane particles. Note that the size of the membrane particles is enlarged for better presentation.

Fig. 13. Flow chart of the numerical simulations.

therefore behaves like the rubber membrane used in the triaxial tests, which is strong enough not to be torn apart but can be stretched to deform freely. Figure 12 illustrates how the target confining pressure can be achieved by adjusting the forces directly applied onto the stress-controlled membrane particles. The Fc represents the equivalent force on the membrane segment, AB, which renders the required confining pressure on the soil particles. Its x- and y-direction components, Fx and Fy, can be calculated based on a geometrical relationship, that is,

y1 y2 Fx Fc p y1 y2 2 x1 x2 2 x1 x2 Fy Fc p y1 y2 2 x1 x2 2

where (x1, y1) and (x2, y2) are the coordinates of particles A and B. The forces, Fx and Fy, are then equally shared by the two membrane particles, A and B that constitute the segment, that is,
#

2008 NRC Canada

36

Can. Geotech. J. Vol. 45, 2008

Fig. 14. Procedures for positioning the cementing particles. Note that the size of the cementing particles is enlarged for better presentation.

Fx Fxa Fxb 2 Fy Fya Fyb 2

Fig. 15. Stressstrain relationships of Portland cement sand obtained from the biaxial-test simulations under 50 kPa confining pressure.

Simulations of biaxial tests

Figure 13 presents the flow chart of the biaxial test simulations and the important details are described in the following sections. Specimen generation The numerical specimens were generated from ten equal layers, and in each layer the sand particle number and the void ratio are identical. Uniform packing with a consistent distribution of local void ratios can be obtained with this method. After the sand particle packing was formed, the cementing particles were positioned following the procedure illustrated in Fig. 14. The specimen was formed without large overlapping and high locked-in forces between particles before the biaxial test was conducted. Biaxial tests The experimental conditions applied to the physical specimen were reproduced in the simulations. The top and bottom caps were modeled with a rigid frictionless plate (or wall). The bottom plate was fixed, and the axial loading was applied only from the top plate. The stress, strain, and overall void ratio were monitored during the test by three major

measurement circles at representative locations in the specimen. Apart from that, thousands of small measurement circles were set to monitor the evolution of local void ratios. The simulations of the biaxial compression tests were divided into two stages: consolidation and shearing. The consolidation was accomplished by applying loading onto the top cap and the membrane particles. The confining pressure was constantly checked and the target value was maintained throughout the test. The shearing stage was strain-controlled so that the top plate moved down at a constant velocity. Static equilibrium was achieved in each strain increment by allowing a very low ratio of unbalanced forces. The total
#

2008 NRC Canada

Wang and Leung

37

Fig. 16. Force-chain distribution of the Portland cement sample under the confining pressure of 50 kPa and axial strain of 1.76%: (a) uncemented sample; (b) 2% cement content sample. Note that the force chain thickness represents the magnitude of contact normal forces.

forces acting on the top and bottom plates were kept at a similar magnitude to ensure equilibrium throughout the specimen.

Fig. 17. Volumetric responses of biaxial-test simulations on Portland cement sand under 50 kPa confining pressure.

Particulate-scale behavior of cemented sand

Enhanced strength As demonstrated in Fig. 15, similar stressstrain relationships, compared with experimental observations, were reproduced for the Portland cement sand in the simulations. The strength enhancement, peak strength, and strain-softening response became more distinct when the cement content was higher. Oda (1972) and Oda et al. (1982) carried out biaxial compression tests on a packing of photoelastic disks. The results demonstrated that the external loading was not evenly shared and only concentrated in the main force-carrying chains. These chains were prone to align with the direction of the major principal stress. Particles outside the main force-carrying chains served as a complementary network to prevent them from buckling upon loading (Radjai et al. 1998; Santamarina et al. 2001; Mitchell and Soga 2005). The buckling events of the main force-carrying chains are responsible for the strength evolution, and, therefore, they detail the failure process (Iwashita and Oda 1998; Radjai et al. 1998; Santamarina et al. 2001; Mitchell and Soga 2005). In this context, the associated underlying mechanisms of the strength enhancement due to cementation can be revealed based on the characteristics of force-chain distributions (i.e., the distribution of contact normal forces). As shown in

Figs. 16a and 16b, the force-chain distributions of the uncemented and cemented sands at an axial strain ("a) of 1.76% are compared. The comparison suggests that all particles in the bonding network jointly share the load and many micro force-chains associated with cementation are generated. The force-chain distribution is relatively uniform and exhibits a webbed pattern in the cemented sand. In addition, the cementation provides an additional support to minimize particle rotation and sliding away from the force chains. Thus, extra energy is required to break the cement bonding to give rise to force-chain buckling. These features lead to a more stable and stronger force-chain complex subjected to smaller
#

2008 NRC Canada

38

Can. Geotech. J. Vol. 45, 2008

Fig. 18. Results of the biaxial-test simulation on 2% Portland cement sand under 50 kPa confining pressure. Note that the deformation characteristics at different strains are also presented by the record of the displacement field.

force concentrations so that higher strength is rendered in the cemented sand. Volumetric dilation As shown in Fig. 17, similar to the experimental observations, simulations revealed that cementation can effectively alter the volumetric response from contraction to dilation, and such a change is enhanced with increasing cement content. The simulation results of the 2% cement content specimen are further analyzed to explore the associated underlying mechanism. Figure 18 presents the relationships between the bondbreakage events and the volumetric and stressstrain responses. The volumetric dilation commences after the sample passes over the initial yield point (i.e., around point F), and the occurrence of bond breakage accelerates afterwards. After the peak strength (i.e., after point G is reached), the particle movement becomes concentrated along the shear

banding direction, which can be seen from the record of the displacement field shown in the same figure. It is worth noting that the severe bond-breakage events occurring between points G and H coincide with great volumetric dilation. This observation implies that the bonded clusters detached from the bonding network assist the volumetric dilation. A further analysis is presented in Fig. 19 where the bonding network, displacement field, force-chain distribution, and contours of particle rotation and local porosity at a strain of "a = 7.15% (i.e., point I in Fig. 18) are jointly considered. A close correlation among these five properties can be established. Intensive bond breakage, concentrated relative particle movement, column-like force chains (instead of a webbed pattern) with force concentrations, great particle rotation, and, most importantly, a high local porosity can be found at similar locations of the sample, especially inside the shear band. In the region outside the shear band, the
#

2008 NRC Canada

Wang and Leung

39

Fig. 19. Features recorded in the biaxial-test simulation on the 2% Portland cement sand under the confining pressure of 50 kPa and the axial strain of 7.15%: (a) bonding network; (b) displacement field; (c) force-chain distribution; (d) particle rotation; (e) local porosity (the initial porosity is 0.2).

bonding network is still preserved and, therefore, the particle rotation and volumetric dilation are prohibited. These findings suggest that monitoring the evolution of these five properties inside the shear band may reveal the underlying mechanism of volumetric dilation in Portland cement sand. The series of illustrations shown in Figs. 2024 and the following discussions are aimed at this revelation. As indicated in Fig. 20, a large void with an elongated shape, void V, was established at "a = 7.15%. Four different types of particles or bonded clusters are involved to form the particle arch that contains this void:
.

Particle C: this is a sand particle coated with cementing particles and functions as a roller to facilitate the sliding of cluster D when void V is formed. Cluster D: this is a properly sized cluster located at the right side of the particle arch.

The formation process of this particle arch and associated void V is described as follows according to the evolution of each participating particle or cluster: (1) Particle A, as shown in Figs. 2124, was originally free to move or rotate and did not participate in the forcecarrying chains at "a = 1.43%. This particle is finally interlocked with particle B to join the left side of the particle arch at "a = 7.15%. Note that particle A is tightly locked by the two cementing particles attached on the surface of particle B, which can be seen in the bonding network and the enlargement at "a = 7.15% (Fig. 24).
#

Particle A: this is a sand particle without any cementation coating Particle B: this is a sand particle coated with cementing particles and serves as a key element to lock the left side of the particle arch

2008 NRC Canada

40

Can. Geotech. J. Vol. 45, 2008

Fig. 20. Descriptions of particles or clusters involved to form void V inside the shear band II at "a = 7.15%.

Fig. 21. Records of displacement field around the region of shear band II.

2008 NRC Canada

Wang and Leung

41

Fig. 22. Records of particle rotation around the region of shear band II. Note that particle B and cluster D have an identical particle rotation at "a = 5.01% but different ones at "a = 5.72%.

(2) Particle B originally belonged to a bonded cluster (i.e., the one that contained cluster D), which can be seen from the bonding network at "a = 1.43% (Fig. 24). This particle was then released and became mobile after the cementation bond was broken at "a = 5.72%. The bond was broken by tensile failure when particle B rotated in a counter direction with respect to its parent cluster. Such a detaching process can be seen by comparing the particle rotation (Fig. 22) and the bonding network (Fig. 24) at "a = 5.01% and "a = 5.72%. This finding explains why the intensive bond breakage and greater particle rotation can be observed at similar locations, as presented in Fig. 19. (3) Particle C was also originally connected to a cluster (i.e., the one that contained cluster D), which, again, can be seen from the bonding network at "a = 1.43% (Fig. 24). The comparisons of the particle rotation (Fig. 22) and the bonding network (Fig. 24) between "a = 3.14% and "a = 5.01% reveal that particle C experienced a similar fate to that of particle B. The cementation bond is subjected to tensile failure when particle C rotates at a different rate with respect to its parent cluster, that is, differential rotation. This released particle later serves as a roller that facilitates the sliding of cluster D as void V is formed. This scenario can be viewed by comparing the particle rotation shown in Fig. 22. Particle C rotates

more than 908 in the clockwise direction when "a increases from 5.72% to 7.15%. (4) As presented in Fig. 24, the bonding network associated with cluster D is almost intact at "a = 1.43%. With increasing strain level, some particles are gradually detached from the bonding network due also to differential rotation and cluster D is therefore formed. At "a = 7.15%, the properly sized cluster D joins the right half of the arch and void V is established. Apart from these scenarios, owing to the great stiffness and strength of Portland cement bonding, the rigid-cluster rotation can also contribute to volumetric dilation, such as that shown by cluster E in Figs. 2024. Note that cluster E is also detached from a large bonding network, as shown in Fig. 24. The last and most important question is how can void V be stabilized to maintain volumetric dilation. As shown in Fig. 23, the force-chain distribution keeps changing with increasing strain. A sudden change in the force-chain distribution takes place, especially in the shear band, when forcechain buckling or bond breakage occurs. These observations, on the one hand, suggest that the force chains established around void V can be destabilized in response to larger strains, which in turn initiates the other life cycle of void formations. On the other hand, these findings imply the im#

2008 NRC Canada

42 Fig. 23. Records of force-chain distributions around the region of shear band II.

Can. Geotech. J. Vol. 45, 2008

portance of cluster D in stabilizing the arch to maintain the large void space for volumetric dilation. With reference to the force-chain distribution at "a = 7.15% shown in Fig. 23, the force chains have a thick column shape on the left side of the particle arch that establishes void V, which indicates a greater force concentration and a higher chance of buckling. On the right side of the arch, the force chains exhibit a webbed pattern because each particle in cluster D jointly shares the load. The risk of force-chain buckling is therefore minimized and void V can be maintained for volumetric dilation. As the experimental results in Fig. 2 demonstrate, the cemented specimens are still subjected to higher deviatoric stress, q, compared with the uncemented specimen at larger strains, for example, at "a = 25%. If a cemented cluster such as cluster D does not help to maintain the stability of the particle arch, which contains open voids, the apparent difference in the void ratios between the cemented and uncemented specimens at this strain level will not be measured (see the volumetric response in Fig. 6). Following this logic, if the cemented cluster is not strong, such an arch with a large void will not exist nor will the volumetric dilation. This partially explains the behavior of the gypsum-cemented specimen, as shown in Figs. 8 and 9.

Summary and conclusion

The mechanical responses of cemented sands not only depend on the density and confining pressure, but they also in-

timately rely on the amount and, in particular, the nature of the cementing agents. Although all of the samples for the triaxial test were prepared in a loose state with very similar initial void ratios (~0.72), the Portland cement (strong cementation) and gypsum-cemented (weak cementation) sand samples exhibited different responses. Increasing the amount of cementation in the Portland cement sand can markedly augment the strength and enhance the associated strain-softening response. The shear banding associated failure mode is noticed. In addition, the Portland cement sand samples, regardless of the cement content, all are prone to exhibit volumetric dilation while the uncemented sample with the same void ratio tends to contract. In the gypsum-cemented sand, the strength enhancement by cementation only prevails under low confining pressure. The strain-hardening behavior and a bulging type of failure are observed and shear banding is not visible. Greater volumetric contraction, if compared with the response of uncemented sand, can be measured. The underlying mechanisms of strength enhancement and volumetric dilation observed in the Portland cement sand are revealed by the aid of DEM simulations. In the simulations, the arrangement of the smaller cementing particles around the contacts between sand grains indicates the real features of cementation, and the use of flexible membrane boundaries improves the simulation of the deformation characteristics. The simulation results demonstrate that all of the
#

2008 NRC Canada

Wang and Leung Fig. 24. Records of the bonding network around the region of shear band II.

43

particles in the bonding network jointly share the load and many micro force-chains associated with cementation are generated. Compared with uncemented sand, a more stable and stronger force-chain network subjected to smaller force or stress concentrations is formed in cemented sand. The risk of force-chain buckling is therefore minimized and higher strength is measured. The simulation also reveals that intensive bond breakage, concentrated relative particle movement, column-like force chains (instead of a webbed pattern) with force concentrations, great particle rotation, and high local porosity can be found at similar locations in the sample, especially inside the shear band. The bonded cluster is essential to help stabilize the particle arch and maintain large voids for the volumetric dilation.

Acknowledgments
This research was supported by the Hong Kong Research Grants Council (HKUST6034/02E) and the Hong Kong University of Science and Technology (HIA04/05.EG02). The authors are grateful to the reviewers for their valuable comments.

References
Abdulla, A.A., and Kiousis, P.D. 1997. Behavior of cemented sands - I. Testing. International Journal for Numerical and Analytical Methods in Geomechanics, 21(8): 533547. doi:10.1002/ (SICI)1096-9853(199708)21:8<533::AID-NAG889>3.0.CO;2-0. Acar, Y.B., and El-Tahir, A.E. 1986. Low strain dynamic properties of artificially cemented sand. Journal of Geotechnical Engineering, 112: 10011015.

Airey, D.W. 1993. Triaxial testing of naturally cemented carbonate soil. Journal of Geotechnical Engineering, 119(9): 13791398. doi:10.1061/(ASCE)0733-9410(1993)119:9(1379). Bass, J.D. 1995. Elasticity of minerals, glasses, and melts. In Mineral physics and crystallography, AGU, reference shelf 2. Edited by T.J. Ahrens. American Geophysical Union, Washington DC. pp. 4563 Clough, G.W., Iwabuchi, J., Rad, N.S., and Kuppusamy, T. 1989. Influence of cementation on liquefaction of sands. Journal of Geotechnical Engineering, 115: 11021117. Clough, G.W., Sitar, N., Bachus, R.C., and Rad, N.S. 1981. Cemented sands under static loading. Journal of Geotechnical Engineering, 107: 799817. Corriveau, D., Savage, S.B., and Oger, L. 1997. Internal friction angles: Characterization using biaxial test simulations. In IUTAM Symposium on Mechanics of Granular and Porous Materials. Cambridge, U.K. 1517 July 1996. Edited by N.A. Fleck and A.C.F. Cocks, Kluwer Academic Publisher, Boston, Mass. pp. 313324. Dupas, J., and Pecker, A. 1979. Static and dynamic properties of sand-cement. Journal of Geotechnical Engineering, 105: 419436. Hazzard, J.F., Young, R.P., and Maxwell, S.C. 2000. Micromechanical modelling of cracking and failure in brittle rocks. Journal of Geophysical Research, 105(B7): 16 68316 697. doi:10. 1029/2000JB900085. Huang, J.T., and Airey, D.W. 1998. Properties of artificially cemented carbonate sand. Journal of Geotechnical and Geoenvironmental Engineering, 124(6): 492499. doi:10.1061/(ASCE) 1090-0241(1998)124:6(492). Ismail, M.A. 2000. Strength and deformation behavior of calcite cemented calcareous soil. Ph.D. thesis, The University of Western Australia, Perth,Australia.
#

2008 NRC Canada

44 Ismail, M.A., Joer, H.A., Sim, W.H., and Randolph, M.F. 2002. Effect of cement type on shear behavior of cemented calcareous soil. Journal of Geotechnical and Geoenvironmental Engineering, 128(6): 520529. doi:10.1061/(ASCE)1090-024 1(2002)128:6 (520). Iwashita, K., and Oda, M. 1998. Rolling resistance at contacts in simulation of shear band development by DEM. Journal of Engineering Mechanics, 124(3): 285292. doi:10.1061/(ASCE) 0733-9399(1998)124:3(285). Kuhn, M.R. 1995. A flexible boundary for three-dimensional DEM particle assemblies. Engineering Computations, 12: 175183. Ladd, R.S. 1978. Preparing test specimens using undercompaction. Geotechnical Testing Journal, ASTM, 1: 1623. Lade, P.V., and Overton, D.D. 1989. Cementation effects in frictional materials. Journal of Geotechnical Engineering, 115: 13731387. Leroueil, S., and Vaughan, P.R. 1990. The general and congruent effects of structure in natural soils and weak rocks. Geotechnique, 40: 467488. Li, X.S., Chan, C.K., and Shen, C.K. 1988. An automatic triaxial testing system. Advanced triaxial testing of soil and rock. In Advanced triaxial testing of soil and rock, ASTM STP 977. American Society for Testing and Materials, Philadephia, Pa. pp. 95106. Mitchell, J.K., and Soga, K. 2005. Fundamentals of soil behavior, 3rd edition. John Wiley & Sons, New York.

Can. Geotech. J. Vol. 45, 2008 Nakata, Y., Hyde, A.F.L., Hyodo, M., and Murata, H. 1999. A probabilistic approach to sand particle crushing in the triaxial test. Geotechnique, 49: 567583. Oda, M. 1972. Deformation mechanism of sand in triaxial compression tests. Soils and Foundations, 12: 4563. Oda, M., Konish, J., and Nemat-Nasser, S. 1982. Experimental micromechanical evaluation of strength of granular materials: effect of particle rolling. Mechanics of Materials, 1: 267283. Pettijohn, F.J., Potter, P.E., and Siever, R. 1987. Sand and sandstone, 2nd edition. Springer-Verlag, New York. Potyondy, D.O., and Cundall, P.A. 2004. A bonded-particle model for rock. International Journal of Rock Mechanics and Mining Sciences, 41: 13291364. doi:10.1016/j.ijrmms.2004.09.011. Radjai, F., Wolf, D.E., Jean, M., and Moreau, J.J. 1998. Biomodal character of stress transmission in granular packing. Physical Review Letters, 80: 6164. doi:10.1103/PhysRevLett.80.61. Santamarina, J.C., Klein, A., and Fam, M.A. 2001. Soils and waves. John Wiley & Sons, New York. Saxena, S.K., Avramidis, A.S., and Reddy, K.R. 1988. Dynamic moduli and damping ratios for cemented sands at low strains. Canadian Geotechnical Journal, 25: 353368. Schnaid, F., Prietto, P.D.M., and Consoli, N.C. 2001. Characterization of cemented sand in triaxial compression. Journal of Geotechnical and Geoenvironmental Engineering, 127(10): 857868. doi:10.1061/(ASCE)1090-0241(2001)127:10(857).

2008 NRC Canada