Advances in Transportation Geotechnics II Miura et al.

(eds) 2012 Taylor & Francis Group, London, ISBN 978-0-415-62135-9

Modelling cemented sand using DEM

J.P. de Bono, G.R. McDowell & D. Wanatowski
University of Nottingham, Nottingham, UK

ABSTRACT: The Discrete Element Method has been used in conjunction with laboratory work to investigate the behaviour of cemented sand under high pressures. DEM simulations of drained triaxial tests with confining pressures up to 12 MPa have been completed, featuring a flexible membrane allowing the correct deformation and failure modes. The inclusion of particle bonding enhances the strength characteristics of the sand, with these effects reducing with increasing confining pressure. Various bond strength distributions have been investigated. The correct qualitative behaviour has been reproduced, with the separate effects of cementation and confining pressures demonstrated. The specimens exhibit very brittle behaviour at lower confining pressures and more ductile behaviour at higher confining pressures as the bonds break prior and during shearing. 1 InTrOductiOn Naturally cemented sand occurs through a number of processes, and as such exhibit high variation in density and degree of cementation (Airey 1993), and there are also difficulties extracting natural specimens while avoiding disturbance. Soils are artificially cemented primarily to improve performance, and are commonly used in subbase layers in pavement construction. Leroueil & Vaughan (1990) showed that structured soils, i.e. cemented sands, over- consolidated clays and weak rocks all follow the same principal behaviour, and that the physical properties of the cemented soil govern its behaviour, rather than the individual cause of cementation. Thus it has been convenient to perform tests on artificially cemented sands to draw conclusions and deepen understanding, for example Clough et al. (1981), Coop & Atkinson (1993), Haeri et al. (2006). Much of the available research using triaxial testing has been carried out at conventional pressures (typically under 1 MPa). Work published by Coop & Atkinson (1993) established that the addition of cementitious material to the sand introduces well defined shear points into the stressstrain response during shearing, which is related to the breakage of cement bonds. After large strains both cemented and uncemented samples reached steady conditions, i.e. critical state and notably appeared to have the same critical state line. They outlined three modes of failure for cemented material, defined by when yielding of the bonding occurs. The first case, bonds yield during isotropic compression, causing shearing behaviour similar to that of the equivalent uncemented soil. The second case, where bonds are initially intact but yield during shearing, and any peak state is governed by the frictional properties of the equivalent nonstructured soil. In the third case, the bonds yield after shearing has begun, causing a clear peak stress to occur. Discrete Element Method (DEM) is a useful tool for modelling granular materials, and has been employed in this study to simulate cemented and uncemented sand in high-pressure triaxial conditions. There has been recent work published using DEM to model and investigate particle bonding (Jiang et al. 2006, Utili & Nova 2008). Wang & Leung (2006) is one of few significant DEM studies of cemented material, although only twodimensional. Belheine et al. (2009) demonstrated that the triaxial test can be successfully modelled in three dimensions using DEM. The principal variables investigated in this study are the numerical bond properties and confining pressure. 2 TriaXial beHaviOur OF cemented sand 2.1 Conventional pressures At low confining pressures, Huang & Airey (1998) demonstrated that cementation causes an increase in stiffness, peak strength, maximum rate of dilation, and in general the specimen becomes more brittle, with these effects all increasing with cement content. For higher confining pressures, the cementation appears broken down and normalised stress-strain responses for various cement contents appear identical.

680

Schnaid et al. (2001) performed a series of drained triaxial tests on cemented sand at low pressures, and demonstrated how largely the soil behaviour is dependent on cement content. They also witnessed an initial volumetric compression followed immediately by a significant expansion, before a steady state was reached, with the higher cement contents exhibiting most dilation. The maximum rate of dilation occurred just after the peak strength, which is dissimilar to uncemented materials, for which the maximum dilation rate coincides with peak deviatoric stress. Similar behaviour has been observed by Ashgari et al. (2003) and Haeri et al. (2004), who also categorised failure modes: brittle failure with shear planes for cemented samples and a barrelling failure mode for uncemented samples. An increase in cement content decreases the strain associated with peak strength. An increase in confining pressure increases the maximum deviatoric stress. It has been shown to cause a transition from brittle to ductile behaviour (Clough et al. 1981, Schnaid et al. 2001, Ashgari et al. 2003). 2.2 High pressures Consolidated drained triaxial tests have been performed on specimens with Portland cement contents from 015% dry weight, by Marri (2010), across confining pressures of 1, 4, 8 and 12MPa. Selected results are displayed in Figure 1. The addition of cement causes an increase in peak deviatoric stress, a reduction in the strain associated with this stress, and an increase in dilation. An increasing amount of cement causes the behaviour to become more brittle. Similar to low pressures, Marri showed that uncemented specimens exhibit strain hardening with no peak deviatoric stress, and undergo

contraction during shearing, demonstrating completely ductile behaviour. Also, the addition of cement causes a peak stress to occur, with the peak stress increasing and becoming more distinguished with higher cement content. With increasing cement content, the axial strain associated with the peak reduces, and the initial modulus of deformation becomes higher. Increasing cement content causes an increase in maximum and rate of dilation. For a given confining pressure, both uncemented and cemented specimens deviatoric stress responses appear to converge or be approaching convergence after large strains, due to the cemented specimens becoming destructured and behaving almost as uncemented. However, they often do not quite converge, especially at high cement contents, due to a portion of bonds remaining intact, effectively changing the grading (Airey 1993). Confining pressure has equally important effects on the behaviour of cemented sand; high pressures suppress the effects of cementation, and render the behaviour from brittle to ductile. Increasing confining pressure causes higher maximum deviatoric stress, and reduces dilation. At the highest pressures, no clear peak stress is apparent, with specimens displaying ductile behaviour, with a gradually increasing deviatoric stress approaching a steady maximum value. The specimens also undergo contraction, in stark contrast to those sheared at lower pressures. However, the effects of cementation can still be observed at higher pressures, when compared with the uncemented material. At intermediate confining pressures, such as 4 and 8 MPa, the behaviour is neither completely brittle nor ductile. The cemented samples at 1 MPa demonstrated brittle behaviour, with clear peak states with strain softening, and shear planes visible in the highly cemented samples. At 12 MPa, all cemented samples demonstrated ductile failure, with gradual strain hardening and volumetric contraction, and barrelling failure modes visible. From lower to higher confining pressures there is a general transition from brittle to ductile behaviour.

3 Discrete element metHOd 3.1 Numerical model of a triaxial test The numerical sample has a height of 100 mm and a radius of 50 mm. The individual sand grains are represented by spherical particles in the software PFC3D. Both cemented and uncemented sand samples are generated using a form of the radii expansion method (Itasca 2005). The current study

Figure 1. Triaxial results for cemented sand under a range of high pressures (Marri 2010).

681

requires the user to specify 3 parameters: void ratio, coefficient of uniformity (D60/D10) and the minimum particle diameter (Dmin). The simplified grading curve is obtained using the coefficient of uniformity (Cu), similar to the method described by Potyondy & Cundall (2004). A value of Cu =2 is used, approximately that of Portaway sand, the base soil used by Marri (2010). Thus for a given void ratio and grading, the specified minimum permitted diameter, Dmin, directly determines how many particles are generated. The number and size of the smallest particle establishes the numerical timestep and calculation time. Varying Dmin will not necessarily change the packing; it shifts the position of the grading curve. Wang & Leung (2006) suggested using a normal particle stiffness of the order of 1*107 N/m for quartz sand, who also suggested the same value can be used for both normal and shear stiffness. In the simulations presented, the sand particle spheres were given normal stiffnesses of 1*107 N/m, and to reduce the number of input parameters, particle shear stiffness was also assigned this value. Density was set at 2650 kg/m3, reflecting the density of the sand used in the above experiments, and a particle friction value of 0.5 was attributed. The specimens were created in a dense initial state, with an initial void ratio of 0.55, consisting of 6759 particles. It is essential that the membrane allows the correct confining pressure to be applied equally to the specimen, while allowing radial deformation. The importance of allowing deformation in threedimensional simulations has been brought to light in recent research such as Cheung & OSullivan (2008); OSullivan & Cui (2009) and Wang & Tonon (2009). However, it is slightly unclear as to how effective these methods were, or if consideration is given to the changing surface area of the specimen. The membrane developed in this study consists of a cylindrical array of bonded particles enclosing the specimen, and a discrete force is applied to each particle combining to give the effects of a confining pressure. The membrane particles are hexagonally arranged. The discrete force applied to each particle is determined from the local surface area of the membrane and the number of membrane particles. The discrete forces are horizontal, and directed towards the central vertical axis of the specimen, and are frequently updated to ensure the confining pressure remains constant while accounting for movement of the membrane. The bonds between membrane particles are Contact Bonds (Itasca 2005), which transmit no moments allowing the membrane to bend. The bonds are given high strengths to avoid breakage. It is important for the membrane particles to be

significantly smaller than the specimen particles, while also taking into account the numerical timestep. Thus the membrane particles used have radii 66% that of the smallest specimen particle. It is not possible to assign realistic stiffness to the membrane particles due to resulting excessive overlap, so higher values are assigned. This results in additional hoop tension applied on the sample, which in turn is relieved by slightly expanding the radii of membrane particles as they bulge outwards. 3.2 Inter-particle bonding Cement bonds are modelled using Parallel Bonds, a feature of the software (Itasca 2005). These consist of a finite-sized piece of material between the two particles, which acts in parallel with the standard force-displacement contact model. These have been used in previous studies, e.g. Potyondy & Cundall (2004). The bonds are defined by: normal/shear stiffness, normal/shear strength and bond size (relative to bonded particles). To reduce the number of variables, bond normal and shear stiffnesses are set equal to the sand particle values. With regards to the literature, it is somewhat unclear how to simulate the size of cement bonds, whether to consider them as small relative to the particles- occurring just at the contacts, independent of particle size; or as proportional to the particles. Both of these approaches seem justifiable depending on interpretation of the nature of cementation. In this study, in order to isolate and investigate bond strength distributions, all bonds are created with equal size (equal to Dmin), therefore having equal moment resistance. The value of bond strength is initially investigated, then an arbitrary value is used to investigate distributions. In the literature, when modelling bonded granular materials, it seems standard practice to install bonds at existing inter-particle contacts. This method, for particles with the density and grading described above, installs approximately 4 bonds per particle, and results in a number of particles with 0 or 1 bonds. Visually inspecting SEM images of cemented sand, such as that in Figure 2 from Marri (2010), immediately suggests a greater number of bonds are required to accurately represent such cementation. The cement can be seen to fill voids and connect particles which would otherwise not be touching. A typical sand particle in the image of 10% cement content in Figure 2 has at least 6 surrounding particles visibly bonded to it, while a sand particle from a denser specimen could have even more. Bearing in mind the planar nature of the image, one would expect a particle not on the surface of a

682

Figure 2. Scanning electron micrograph of portaway sand with 10% cement content.

The peak stress evident in the two strongest samples is caused by the presence of bonds, with the peak coinciding with the onset of major bond breakage. All bonded samples exhibit a much stiffer initial response compared to the unbonded sample. The cemented samples approach the stress state of the uncemented sample as the bonds eventually become broken down, as the material becomes destructured. The bond uniformity is reflected in the graphs, with the very sharp peak indicating extremely brittle failure and sudden breakage. It is apparent that there is an initial linear region, during which no bonds have yielded. This seems somewhat inaccurate, especially at these stress levels. The figures compare three distributions: Uniform, Normal, and Weibull. A Weibull distribution is defined by two variables, the modulus, m, which determines the shape of the distribution, and a parameter, which determines the scale. The cumulative distribution function for a Weibull distribution is given by:
x F ( x; m; ) = 1 e ( )
m

Figure 3. Triaxial response for samples with uniform bonds.

specimen to be bonded to an even greater number, such as 1020. The simulations in this section solely investigate bond strength distributions, and considering the numerical sample has an approximately equal coefficient of uniformity and density to the material in the figure, approximately 33,000 bonds are installed in the following simulations, meaning each particle has an average of 10 neighbouring particles bonded to it. This is achieved by bonding particles which arent necessarily touching, but lie within a specified proximity of one another. In addition to the cemented simulations, a sample with no bonds representing uncemented sand is included. The principal effects of bonds are evident in Figure 3, where the results are displayed from the simulations of an uncemented sample, and three cemented samples sheared under a confining pressure of 4 MPa. Each cemented sample has uniform bond strength, i.e. no distribution. All bonds are equal in size, stiffness and strength in each test, with the (mean) strength varied across the three simulations. It is immediately evident that the introduction of cement bonds with strengths 50 and 100 N cause a peak deviatoric stress to appear, and hence cause a much greater maximum stress compared to the unbonded sample. The simulation with bonds of strength 25N does not have a clear peak, with almost all the bonds breaking soon upon the beginning of shearing, after which behaviour similar to the uncemented sample is witnessed.

And the mean of the distribution is given by: 1 = 1 + m Where G is the gamma function. Changing the modulus (shape) of a Weibull distribution changes its range as well as slightly altering the mean, so the scale of the distribution needs to be adjusted to maintain the mean. For a given scale (), increasing the modulus results in a narrower distribution. Weibull probability is used widely in materials science, especially in failure probability of brittle materials, so it seems reasonable that such a distribution of may be applicable to cementation. It has also been used in the field of geotechnics, in particular particle breakage, for example McDowell & Amon (2000) and McDowell (2001). Figure 4 shows the triaxial results for 2 specimens with uniform bond strength distributions and 2 with normal distributions. The uniform distributions have ranges of 50 and 100N and normal distributions have coefficients of variation (Cv) of 0.2 and 0.4. Figure 5 displays the results of 4 simulations with bond strengths from a Weibull distribution, with varying modulus. The sample with a uniform distribution of bond strengths ranging from 2575 N, appears very similar to the simulation with no distribution. It has a finite minimum bond strength of 25 N, so there will still be an initial linear region, during which no bond breakage is witnessed. The sample

683

Figure 4. Triaxial behaviour for various bond strength distributions.

Figure 6. Bonded sample sheared at various confining pressures (approximately 10 bonds per ball).

in Figure 1. This indicates an earlier, less sudden onset of breakage. Decreasing the shape parameter renders the behaviour more ductile. Although a number of bonds are broken during consolidation, the onset of breakage is much more gradual. 3.3 Modelling various confining pressures Using a Weibull bond strength distribution, with a shape parameter of 0.5, triaxial simulations have been performed over a range of high confining pressures. Assuming that the above bond parameters are representative of a given cement content, Figure 6 shows the effects increasing the confining pressure has on the deviatoric stress. Figure 6 shows the results from simulations with approximately 10 bonds per particle conducted at confining pressures of 1, 4, 8 and 12 MPa. It is evident from this graph that increasing the confining pressure leads to a higher maximum deviatoric stress. The graph is comparable with experimental results in Figure 1. The strain associated with the maximum deviatoric stress increases with confining pressure. The peak is much more prominent at lower pressures, becoming less distinguished at 12 MPa. As with experimental results, there is a clear transition from brittle to ductile behaviour, with the effects of cementation/bonds being suppressed by increasing confinement. This change in failure behaviour from brittle to ductile is also visible in the sheared samples. Figure 7 shows the particle rotations at 1 and 12MPa confining pressures. Dark signifies particles which have undergone the most rotation; white the least. No clear shear plane is visible at 12 MPa, while there are conjugate shear planes visible at 1 MPa. The samples sheared at higher pressures display the correct barrelling failure and in between there is transitional behaviour. Increasing confining pressures results in a larger portion of bonds broken both before and during shearing.

Figure 5. Triaxial behaviour for various weibull distributions.

with bond strengths ranging from 0100N exhibits a slightly less distinct peak stress. The specimen experiences some bond breakage during consolidation and immediately upon shearing, however there is still a sharp peak stress. The samples with normal bond strength distributions give very similar stress-strain results, despite exhibiting slightly different bond breakage. The sample with a coefficient of variation of 0.2 tends to behave similarly to the sample with no bond distribution. The sample with a coefficient of variation of 0.4 exhibits slightly earlier onset of breakage, although the peak stress again appears sharp, indicating sudden failure, due to most bond strengths lying close to the mean value. Finally, the graph displaying Weibull distributions shows that the lower the modulus the more rounded the peak stress state. Higher values render the distribution and behaviour of the bonded sample similar to that of a normal distribution. Lower values (i.e. 1) produce a positively skewed, very wide distribution of strengths. It can be seen that the sample with m = 0.5 produces a very rounded peak stress, which appears similar to the stress-strain curves at high pressures

684

Figure 7. Particle rotations of bonded samples sheared under 1 MPa (left) and 12 MPa (right), at 3% strain (approx. 10 bonds per ball). Darker shade indicates larger rotation.

4cOnclusiOns A model has been created using DEM which successfully simulates high pressure triaxial tests, with a flexible membrane allowing the correct deformation. Several simulations investigating the behaviour of cemented sand have been carried out, principally investigating bond strength distributions, and how bonds behave under various confining pressures. It has been shown that a Weibull distribution of bond strengths with a low shape parameter/ modulus produces the most realistic appearing deviatoric stress response for cemented sand. From analysing microscopic images of cemented sand, cement content has been related to the number of bonds in the numerical sample. Using a Weibull distribution of bond strengths, the correct qualitative behaviour for cemented sand has been witnessed in the simulations with regards to varying confining pressure. The correct transition from brittle to ductile behaviour is witnessed with increasing confining pressure, with high pressures suppressing the effects of cementation. reFerences
Airey, D. 1993. Triaxial Testing of Naturally Cemented Carbonate Soil. Journal of Geotechnical Engineering, 119(9): 13791398. Ashgari, E., Toll, D. & Haeri, S. 2003. Triaxial Behaviour of a Cemented Gravelly Sand. Geotechnical and Geological Engineering, 21: 128.

Belheine, N., Plassiard, J., Donze, F., Darve, F. & Seridi, A. 2009. Numerical Simulation of Drained Triaxial Test using 3D Discrete Element Modelling. Computers and Geotechnics, 36: 320331. Cheung, G. & OSullivan, C. 2008. Effective Simulation of Flexible Lateral Boundaries in two and three- dimensional DEM Simulations. Particuology, 6: 483500. Clough, G., Sitar, N., Bachus, R. & Rad, N. 1981. Cemented Sands under Static Loading. Journal of Geotechnical Engineering, 107: 799818. Coop, M. & Atkinson, J. 1993. The Mechanics of Cemented Carbonate Sands. Geotechnique, 43(1): 5368. Haeri, S., Hamidi, A. & Tabatabaee, N. 2004. The Effect of Gypsum Cementation on the Mechanical Behaviour of Gravely Sands. Geotechnical Testing Journal, 28 (IV). Haeri, S., Hamidi, A., Hosseini, S., Asghari, E. & Toll, D. 2006. Effect of Cement Type on the Mechanical Behaviour of a Gravely Sand. Geotechnical and Geological Engineering, 24(2): 335360. Huang, J. & Airey, D. 1998. Properties of Artifically Cemented Carbonate Sand. Journal Geotechnical and Geoenvironmental Engineering, 124(6): 492499. Itasca, 2005. Particle Flow Code in Three Dimensions, Software Manual. Minnesota, MN: Itasca Consulting Group Inc. Jiang, M., Yu, H.-S. & Leroueil, S. 2006. A Simple and Efficient Approach to Capturing Bonding Effect in Naturally Microstructured Sands by DEM. International Journal of Numerical Methods in Engineering, 69: 11581193. Leroueil, S. & Vaughan, P. 1990. The General and Congruent Effects of Structure in Natural Soils and Weak Rocks. Geotechnique, 40(3): 467488. Marri, A. 2010. The Mechanical Behaviour of Cemented Granular Materials at High Pressures (PhD Thesis). University of Nottingham. McDowell, G. 2001. Statistics of Soil Particle Strength. Geotechnique, 51 (10): 897900. McDowell, G. & Amon, A. 2000. The Application of Weibull Statistics to The Fracture of Soil Particles. Soils and Foundations, 40(5): 133141. OSullivan, C. & Cui, L. 2009. Micromechanics of Granular Material Response during Load Reversals: Combined DEM and Experimental Study. Powder Technology, 193: 289302. Potyondy, D. & Cundall, P. 2004. A Bonded-Particle Model for Rock. International Journal of Rock Mechanics & Mining Sciences (41): 13291364. Schnaid, F., Prietto, P. & Consoli, N. 2001. Characterization of Cemented Sand in Triaxial Compression. Journal of Geotechnical and Geoenvironmental Engineering, 127(10): 857868. Utili, S. & Nova, R. 2008. DEM Analysis of Bonded Granular Geomaterials. International Journal for Numerical and Analytical Methods in Geomechanics, 32: 19972031. Wang, Y. & Tonon, F. 2009. Modelling Triaxial Test on Intact Rock using DEM with Membrane Boundary. Journal of Engineering Mechanics, 135(9): 10291037. Wang, Y.-H. & Leung, S.-C. 2006. A Particulate-Scale Investigation of Cemented Sand Behvaiour. Canadian Geotechnical Journal, 45: 2944.

685