This action might not be possible to undo. Are you sure you want to continue?
NUMERICAL METHODS IN GEOMECHANICS
School of Civil Engineering, Purdue University, West Lafayette, IN, USA
ﺗﻘﺪم هﺬﻩ اﻟﻮرﻗﺔ وﺻﻔﺎ ﻟﻠﻄﺮق اﻟﻌﺪدﻳﺔ اﻷآﺜﺮ اﺳﺘﺨﺪاﻣﺎ ﻓﻲ اﻟﻤﻴﻜﺎﻧﻴﻜﺎ اﻟﺠﻴﻮﻟﻮﺟﻴﺔ. وهﻲ أرﺑﻌﺔ ﻃﺮق : )1( ﻃﺮﻳﻘﺔ اﻟﻌﻨﺼﺮ اﻟﻤﻤﻴﺰ )2( ﻃﺮﻳﻘﺔ ﺗﺤﻠﻴﻞ ً .اﻟﻨﺸﻮة اﻟﻤﺘﻘﻄﻊ )3( ﻃﺮﻳﻘﺔ اﻟﺘﺤﺎم اﻟﺠﺴﻴﻤﺎت )4( ﻃﺮﻳﻘﺔ اﻟﺸﺒﻜﺔ اﻟﻌﺼﺒﻴﺔ اﻟﺼﻨﺎﻋﻴﺔ ً ً ً .وﺗﻀﻤﻨﺖ اﻟﻮرﻗﺔ أﻳﻀﺎ وﺻﻔﺎ ﻣﻮﺟﺰا ﻟﺘﻄﺒﻴﻖ اﻟﻤﺒﺎدئ اﻟﺨﻮارزﻣﻴﺔ ﻟﻜﻞ ﻃﺮﻳﻘﺔ إﺿﺎﻓﺔ إﻟﻰ ﺣﺎﻟﺔ ﺑﺴﻴﻄﺔ ﻟﺘﻮﺿﻴﺢ اﺳﺘﺨﺪاﻣﻬﺎ
______________________ *Corresponding Author: E-mail: email@example.com Paper Received November 7, 2009; Paper Revised January 17, 2009; Paper Accepted February 3, 2010
The Arabian Journal for Science and Engineering, Volume 35, Number 1B
ABSTRACT The paper presents a description of the numerical methods most used in geomechanics. The following methods are included: (1) The Distinct Element Method; (2) The Discontinuous Deformation Analysis Method; (3) The Bonded Particle Method; and (4) The Artificial Neural Network Method. A brief description of the fundamental algorithms that apply to each method is included, as well as a simple case to illustrate their use. Key words: numerical methods, geomechanics, continuum, discontinuum, finite difference, finite element, discrete element, discontinuous deformation analysis, bonded particle, artificial neural network
The Arabian Journal for Science and Engineering, Volume 35, Number 1B
The method really took off with the advent of computers. continuum or discontinuum methods depends on the size. Discontinuous Deformation Analysis (DDA). Such assumptions usually include elastic behavior. either the medium and the boundaries (FEM and FDM) or only the boundaries (BEM) are discretized. in this case the displacement field is continuous and thus continuum numerical methods are appropriate. of the problem that needs to be solved. numerical methods can be classified into continuum and discontinuum methods [1. while in discontinuum methods. The following sections provide the fundamental assumptions and the mathematical framework for each method and an overview of the range of problems where each method has been successfully used. Numerical methods have been extensively used in the past several decades due to advances in computing power. Figure 1(a) illustrates an opening in a medium without discontinuities. Rocks and soils may not be isotropic or homogeneous. past stress history). The displacement field will be continuous inside each region but may be discontinuous across the discontinuities. The method is the oldest among the numerical methods in geomechanics and was used even before the arrival of computers. April 2010 The Arabian Journal for Science and Engineering. INTRODUCTION Analytical methods are very useful in geomechanics because they provide results with very limited effort and highlight the most important variables that determine the solution of a problem. This is so because some simplifications are made to solve the system of differential equations either inside the continuum or at the boundaries of the discretization. Discontinuum.2]. The need to use. Finite Element Method (FEM) and Boundary Element Method (BEM). either because this is inherent to the material or because it has been externally induced (e. DDA. and the Boundary Element Method (BEM) are the continuum methods most used in geomechanics [4. Numerical methods give only approximations to the correct or exact mathematical solution. The medium depicted in Figure 1(c) is determined by a number of discontinuities with spacing and continuity such that the blocks defined are within the scale of the opening. this promoted a much wider use of the FDM.g. or at least the most used methods are: Continuum. In this case. Analytical solutions. the model should be able to consider the specific discontinuities.5]. There are no quantitative guidelines to determine when one method should be used instead of the other one.. It has to be mentioned also that the problem that is solved is the conceptualization that is done of the actual physical problem. or the geometry of the problem may be complex. who in 1908 used it for the solution of torsion problems. In a broad sense. for a particular problem. There are two other methods which do not follow this classification: Meshless Methods (MM) and Artificial Neural Networks (ANN). If a continuum model is used. homogeneous. AND BOUNDARY ELEMENT METHOD The Finite Element Method (FEM). BPM. and Bonded Particle Model (BPM). Southwell  developed the relaxation method. Thus. explicit or implicitly. a pseudocontinuous displacement field is produced and the use of a continuum model seems reasonable. solutions can only be obtained numerically. which have recently seen a significant use growth.g.. A brief description of continuum methods is also included. In these methods. The conceptualization applies to the geometry of the problem. the paper focuses on DEM. isotropic material. Number 1B 29 . 2. Figure 1(b) shows a tunnel excavated in a medium with a small number of discontinuities which divide the medium into a small number of continuous regions. however. There is quite a large number of numerical methods that have been used in the literature to estimate the behavior of geomaterials. discontinuities are incorporated explicitly. If the medium is heavily jointed such that the blocks defined by the discontinuities have a size much smaller than the opening. and ANN. which can be solved by any of the classical methods. the Finite Difference Method (FDM). FINITE ELEMENT METHOD. which provides a fast solution of the system of equations. etc. Distinct Element Method (DEM). Figure 1 (following Brady ) provides some qualitative guidance. of the discontinuities with respect to the size. and the response of the geomaterials to loading. if present. have often a limited application since they must be used within the range of assumptions made for their development. The Finite Difference Method (FDM) is based on the premise that governing differential equations can be adequately represented by finite differences. the set of differential equations is reduced to a system of linear equations. and the loading may not be static. In these cases. or scale. quasi-static loading. Finite Difference Method (FDM). Continuum methods may incorporate the discontinuities in the medium. displacements may be determined by the slip along the discontinuities and rotation of the blocks. For example.Antonio Bobet NUMERICAL METHODS IN GEOMECHANICS 1. a discontinuum numerical method seems appropriate. With the FDM. Geomaterials such as soils and rock masses display non-linear behavior. CONTINUUM METHODS: FINITE DIFFERENCE METHOD. time independent behavior. While all methods are relevant. Figure 1(d). e. the loading process or history. the more accurate the solution will be with respect to the response observed in the field. Timoshenko and Goodier  attribute the first application of the method to Runge. or scale. The better the approximation to the field problem through this conceptualization process. The most important. Volume 35.
xj).g. equilibrium. and frictional laws (e.j the point (xi+h. xj). Finite Difference Grid in 2D 30 The Arabian Journal for Science and Engineering. Figure 2. The relative displacement between corresponding grid points determines the slip along the discontinuity. for example. xj+k). In the figure.e. etc. the sub-indices represent the position of the point in the grid. Number 1B April 2010 .Antonio Bobet Figure 1.. strain compatibility. i. by finite difference formulas. and so on. Coulomb) can be enforced by adding new equations to the system of equations that relate shear stress with normal stress. Discontinuities can be incorporated in the model by using grid points on each side of the discontinuity.j+1 the point (xi. i+1. i.j represents a point with coordinates (xi. as shown in Figure 2. The method relies on the approximation of the field equations. Normal and shear displacements can also be related to the shear and normal stiffness of the discontinuity. Volume 35. Continuum and Discrete Models With the method a grid is superimposed to the domain. i.
where the entire medium has to be discretized. and ρ the density of the medium. Number 1B 31 . At the end of each loading step. Finite Element Discretization in 2D With the Boundary Element Method (BEM). displacements at the grid points are obtained. This is in contrast to the other two continuum methods. While the number of steps is quite large. and. The method relies on the assumption that. the solution is approximated at the boundaries while equilibrium and compatibility are exactly satisfied in the interior of the medium. Dynamic problems require a maximum time step to ensure stability of the solution. Also. satisfies boundary conditions) small virtual displacements applied to the body. and from 2D to a line problem. ∆x is the grid spacings. which is common in problems in geomechanics. G the shear modulus. Figure 3. the Finite Difference and the Finite Element methods. and another small loading increment is added. The new increment starts with the updated stress field from the previous increment. unlike other techniques such as Finite Element Methods that use implicit solution methods. complex dynamic problems can be analyzed in a reasonable period of time. if the medium extends to infinity. through appropriately chosen interpolation functions. The BEM automatically satisfies far-field conditions. Equation (1) indicates that the maximum time step is controlled by the stiffer material in the medium. K is the bulk modulus. Thus the method is very attractive for those problems where the volume to boundary surface ratio is large. The solution is then obtained on a stepwise process involving sufficiently small loading increments until the desired final state is reached. This is a forward scheme that does not require iteration. April 2010 The Arabian Journal for Science and Engineering. displacements at any point within the element can be accurately obtained from the displacements of the nodes. any compatible (i. for a body in equilibrium. where displacements are a function of position and time. In FDM and FEM. In the BEM. The method consists of discretization of the continuum into small elements that intersect at their nodes (Figure 3).e. the time required to complete each step and the memory required to store the solution is small. no artificial boundaries such as those needed in FDM and FEM are required. See Figure 4. which is given by ⎛∆ x⎞ ⎟ ∆t = min ⎜ ⎜ Cp ⎟ ⎝ ⎠ Cp = K+4/3 G (1) ρ Cp is the compressional or P-wave velocity in the medium. according to Bathe .. The term “Finite Element”. The method is based on the principle of virtual displacements. was first introduced by Clough . only the boundaries of the continuum need to be discretized. The advantage of limiting the discretization to the boundaries is that the problem is reduced by one order: from 3D to a 2D surface problem at the boundary. Explicit time integration techniques are often used to provide solutions using small time increment steps. the approximations are made inside the medium. Volume 35. thus.Antonio Bobet The method can also be readily used to solve dynamic problems. stresses are then updated based on the non-linear behavior of the material. It is not unusual to run tens of thousands of steps to complete a numerical analysis. which states that. The finite difference approach is very well-suited to incorporate non-linear behavior. the total internal work associated with the virtual displacement field must be equal to the total virtual external work. The Finite Element Method (FEM) is by far the method used the most for the analysis of continuous or quasicontinuous media.
Boundary Elements are used. where a FEM discretization is used for those parts of the continuum where plastic deformations occur. typically stresses or displacements. which Cundall and Hart  define as those that: (1) allow finite displacements and rotations of discrete bodies. blasting. Efficient hybrid BEM-FEM solutions are possible. 32 The Arabian Journal for Science and Engineering. and (2) automatically recognize new contacts between bodies during calculations. In the first approach (Direct BEM). Far from the tunnel and where the deformations are elastic. Number 1B April 2010 . and with stresses or displacements applied to the boundaries. The advantage of the coupled FEM-BEM is reduced discretization and automatic satisfaction of boundary conditions at infinity. to integral equations which only consider boundary values [10–12]. In a boundary value problem.15–19] for both soils and rocks. and once they are obtained. are discretized with Finite Elements. Actual problems may not always conform to these limitations. Volume 35. In the second approach (Indirect BEM). and inertia may play an important role with dynamic loading (e. which apply to the entire medium. and that the resulting stiffness matrix is non-symmetric (in contrast with FEM where the stiffness matrix is generally symmetric). 3. With plastic deformations and with cases where integration needs to be extended over part or the entire volume. and then used for soils which were modeled as discs .. where the tunnel liner and a volume of the ground next to the tunnel where plastic deformations occur. Discrete Element Methods need to address three key issues: (1) representation of contacts. Figure 5 shows an example of a hybrid discretization of a tunnel. including detachment. earthquake). the solution is found in terms of some “fictitious” quantities. is attained through iteration. For example. gravity forces may be significant for shallow geostructures. Boundary Element Methods are particularly well-suited to address static continuum problems with small boundary to volume ratios. the advantage that the BEM offers regarding limited discretization of the continuum may be lost. and (3) detection and revision of contacts during execution.Antonio Bobet Figure 4. some parameters such as stresses and displacements are known while others are not. as with FEM. stresses and displacements at any point in the continuum can be obtained directly from the solution. the unknowns are solved directly. Later on. An in-depth discussion of these issues is provided by Cundall and Hart .g. the method has been applied to spherical and polyhedral blocks [4. The challenge of the hybrid approach is the generation of nodal forces and displacements from the BEM that are consistent with those of FEM. (2) representation of solid material. The DEM belongs to the family of Discrete Element Methods. The plasticity algorithms require integration at least over the volume of the material that undergoes yielding and convergence of the solution. The fictitious quantities are obtained first and the stresses and displacements at any point in the medium are expressed in terms of these fictitious quantities. There are two approaches to solve for the unknown parameters. Dynamic and body forces require integration over the entire volume domain which leads to the need for discretization of the entire continuum. DISCONTINUUM: THE DISTINCT ELEMENT METHOD The Distinct Element Method (DEM) was introduced by Cundall  as a model to simulate large movements in blocky rock masses. Example of Discretization with Boundary Elements in 2D The technique used in BEM consists in essence of transforming the governing differential equations. while Boundary Elements are used in elastic regions. with elastic behavior. which then are part of the solution. rocks and soil deposits may undergo significant yielding under moderate stresses.
Volume 35. the discussion is restricted to two-dimensional discretizations with rigid bodies. The displacement of the block is governed by Newton’s second law of motion: m u t + c u t = Ft i i i t + c ωt = M t I ω (2) April 2010 The Arabian Journal for Science and Engineering. if any. gravity).g. Figure 6. Hybrid BEM – FEM Mesh In the DEM. which in turn are interconnected through the discontinuities. Distinct Element Method Discretization A single block is subjected to forces arising from the contacts. For clarity. The following provides key concepts for the formulation of the DEM. it is assumed that the medium is divided by fully persistent discontinuities which delimit through their intersections a finite number of blocks. Number 1B 33 .Antonio Bobet Figure 5.. Figure 6 provides an idealization of a discretization with DEM of a medium with two sets of discontinuities. from the surrounding blocks and from internal forces (e.
ui is the displacement of the gravity center of the element in the direction i. and Fi and M are the resultant force and moment applied at the center of gravity. a normal and a shear force appear as the result of the relative movements between the two elements that share the discontinuity. In the DEM Equation (2) is solved in the time domain using an explicit finite difference method. I is the moment of inertia of the element. Forces at the Boundary of DEM Elements 34 The Arabian Journal for Science and Engineering. ∆un and ∆us are relative displacements between the two elements.g. At each boundary. Number 1B April 2010 . m is the mass. The forces at the interface may be obtained using a penalty method where the magnitude of the forces is related to the relative movements between the two elements and the stiffness of the discontinuity. respectively). approximation velocities and displacements are given by ⎡ t ⎤ t +∆t / 2 = ⎢ D u t −∆t / 2 + Fi ∆t ⎥ D u ⎢ 1 i ⎥ 2 i m ⎢ ⎣ ⎥ ⎦ M ωt +∆t / 2 = ⎢ D1 ωt −∆t / 2 + ∆t ⎥ D 2 ⎢ ⎣ m ⎥ ⎦ c ∆t m 2 1 D2 = c ∆t 1+ m 2 D1 = 1⎡ t ⎤ (3) i i i t +∆t = θ t +ωt +∆t / 2 ∆t θ The forces acting at the boundaries are originated by the interaction of the element with the surrounding elements.β K n ∆u ∆t Ac n n t t ∆ ∆ Fs +∆t = Fs . and Figure 7(b) shows an idealization of the contact between blocks. damping proportional to the rate of change of the kinetic energy of the element ). but other expressions for damping have been proposed (e. c is the viscous damping. u i and ui are the acceleration and the velocity of the gravity center. the third term on the right-hand side of the equation.Antonio Bobet where t is time.. ω and ω are the angular rotation and angular velocity of the element.β Ks ∆u s t Ac (4) ut +∆t = ut +u t +∆t / 2 ∆t Kn and Ks are the normal and shear stiffness of the contact (subscripts n and s refer to the directions normal and parallel to the discontinuity. Damping (Cn and Cs in Figure 7) is often expressed as proportional to the normal and shear stiffness (βKn and βKs in (4)). The shear force is proportional to the relative movement along the direction of the contact. is normally included to attenuate or prevent “rattling” of the contact between blocks.Ks ∆u s t Ac .K n ∆u ∆t Ac . Volume 35. Figure 7. Using the central finite difference. and Ac is the contact area. Expressions for the forces are t t Fn +∆t = Fn . The normal force is proportional to the relative movement of the two blocks across the contact and along the normal direction. A damping factor. Figure 7(a) shows the positive forces at the top of the element.
As a result.27]. Therefore. The simulations were run using parallel processing and the Livermore Distinct Element Code (LDEC). A very simple procedure would be to compare the position of each element with the rest of the elements at the end of each cycle. making these the largest simulations of this type performed to date. The calculations are performed from one state. and cohesion) with the DEM approach. Figures 8(b) and (c) show the response of the tunnel immediately after detonation and 30 ms later. contact forces are updated using Equation (4) and new resultant forces and moments at the center of gravity of each element are computed. and consisted of 8 million blocks with approximately 100 million contacts. Figure 8(a) shows the discretization used to investigate the response of a tunnel in a discontinuous rock mass subjected to blast loading [30. The procedure is as follows : The law of motion is applied through Equation (3) with current forces to update the position of each element. the relative displacements and velocities at the contacts between elements are obtained.25] and to fragmentation of discontinua [20. which would make the entire method impractical. friction angle. this would require of the order of n2 operations in each cycle. or a fictitious parameter to represent loading increment from one loading stage to the next. April 2010 The Arabian Journal for Science and Engineering. and on the other hand are not computationally intensive. Kmax is the largest normal or shear stiffness in the discretization. since the magnitude and direction of the contact forces depend on the type of contact. A number of approaches has been proposed to identify contacts. A value for κ equal to 0. The problem is complex as the algorithms need to identify not only what elements are in contact but also the type of contact: corner to corner.Antonio Bobet The magnitude of the shear force is limited by the constitutive relation used for the contact surface. and the joints are allowed to undergo large deformations. For a discretization with n elements. which on the one hand need to accurately describe the interaction between elements. and κ is a factor that takes into account the fact that an element may be in contact with more than one element.18. In the formulation. The cycle is repeated with small increments until the final solution is obtained. For a Coulomb-type friction law. buffer zone definition. binary tree structures. The solution of equations (3) and (4) is a forward process.1 has been suggested . From the relative displacements. the storage information needed for each element is small. Volume 35. One fundamental advantage of the DEM is that pre-existing joints in rock can be incorporated into a DEM model directly. Typical runs are completed with thousands of cycles involving very small time increments. t t Fs +∆t ≤ c Ac + Fn +∆t tan φ (5) where c and φ are the cohesion and friction angles of the contact surface. and. can be applied to deformable bodies [4. Heuze and Morris  provide an extensive overview of the DEM as applied to jointed rock masses. which is given by : ∆tcrit = κ mmin 2 K max (6) where mmin is the smallest element mass. contact or field zone. the computer time required in each cycle is very small. It can be used for static and for dynamic calculations [26. with typical block size of 30 cm. or edge to edge. time can represent actual time when performing a dynamic analysis. to another state in small time increments. If the shear force obtained from (4) is larger that that from (5). such as global searching algorithms. Where such requirements become significant is for the algorithm to recognize and keep track of all the contacts between elements during execution. Number 1B 35 . corner to edge.20–23]. Detailed joint constitutive models (see  for a review) can also be used to combine experimentally observed fracture properties (such as joint dilation. numerical stability requires a time increment smaller than the critical time step. also. and by discretizing the elements with Finite Difference or Finite Element meshes. the process discussed so far does not require intensive computation power or large storage capabilities. The Distinct Element Method is nowadays a very versatile and extensively validated procedure. it is reduced to the limiting magnitude given by (5). It has been developed for full three-dimensional problems. 25]. thus. As with the Finite Difference Method. space decomposition and alternating digital tree [16. A comprehensive review of these methods is provided in . where the solution is fully known. Considerable effort has been done to develop efficient algorithms.31].
yo and ai and bi i=1. Within each step. The method started with the work of Shi and Goodman [32.2 are constants. and their shapes are determined by the location of their contacts with the neighboring elements. edge to edge.Antonio Bobet Figure 8. they can be given. Volume 35. by a first order approximation of the form u = u o + (x-x o ) a1 +(y-yo ) a 2 v = vo + (x-x o ) b1 +(y-yo ) b2 (7) where u and v are the x. DEM simulation of an underground structure subjected to dynamic loading (Morris and Block. There are fundamental differences between the DEM and DDA. The DEM uses an explicit procedure to solve the equilibrium equations and the DDA is an implicit method. in 2-D. (2) representation of solid material. with the contacts between blocks consisting of edge to face. While the DDA is a fully discontinuous analysis. while in the DDA.35]. It is assumed that any large displacements or deformations are the result of the accumulation of small displacements and deformations after a sufficiently large number of steps. interpenetration of blocks is prevented by adding springs to the contacts. needs to address three key issues: (1) representation of contacts. or face to face. In turn displacements can be expressed as a function of strains as follows: 36 The Arabian Journal for Science and Engineering. similar to the DEM.and y-axis displacements of a point with coordinates x and y. while in the DDA. DISCONTINUUM: DISCONTINUOUS DEFORMATION ANALYSIS The Discontinuous Deformation Analysis (DDA) is a Discrete Element Method following the definition by Cundall and Hart . stresses and forces are unknowns while displacements are computed from stresses. as outlined in the preceding section. thus. Strains can be computed from (7). Thus. in DDA the displacements are the unknowns. the total potential energy of the system is minimized to find the solution. the medium is discretized into elements or blocks which are in contact with each other only through the discontinuities. In essence. Number 1B April 2010 . The elements can be convex or non-convex. In the DEM. it resembles and follows the procedures developed for FEM. uo and vo are the rigid body motions at point xo. and (3) detection and revision of contacts during execution. The DDA.33]. The method is fully described in [34. In the DEM. and since then it has received considerable attention by the geoengineering community. the displacements of any block are small and. In the DEM each block is treated separately. 2006) 4. The discretization used in Figure 6 to illustrate the DEM could perfectly apply to the DDA. blocks are represented by polyhedra. the contacts are resolved using a penalty method which results in the definition of the contact forces.
Kii depends on the material properties of element i and Kij (i≠j) on the contacts between elements.Antonio Bobet ⎛1 ⎞ u = u o + (x-x o ) ε xx +(y-yo ) ⎜ γ xy -ro ⎟ ⎝2 ⎠ 1 ⎛ ⎞ v = vo + (y-yo ) ε yy +(x-x o ) ⎜ γ xy +ro ⎟ ⎝2 ⎠ (8) where εxx. from initial conditions. with other numerical methods. Figure 9(a). and adding stiff springs is repeated until no interpenetration occurs. etc. initial stresses. the relative displacement between two elements is smaller than their initial distance. Figure 9(f). and Fi is a set of 6x1 force submatrices of element i. Note that strains in each element are constant. concentrated and distributed loads. where blocks are divided into sub-blocks when tensile or shear stresses reach the strength of the material. thus. Figure 9 illustrates an example application of the DDA method . If within a single step. once the lock is removed (i. determining interpenetration. is expressed as K ij D j = Fi (10) Dj is made of 6x1 sub-matrices that contain the 6 unknowns of each element j. then stiff springs are placed between the two elements and the system is recalculated. a spring normal to the contact is placed to allow for sliding and prevent penetration in the normal direction. potential contacts between elements are updated. The method. [40–45. in radians. by adding or removing locks along appropriate directions. Kij is composed of 6x6 stiffness sub-matrices associated with the corresponding degrees of freedom of element j. If the component of the force parallel to the contact is larger than the maximum allowed by the constitutive model (e. Within a certain loading step (load increment). and (2) add to the global equilibrium equations the appropriate stiff springs.33. a shallow rectangular tunnel in a rock mass medium with two joint sets is subjected to a vertical load on the surface. thus preventing penetration. In the DDA. As the simulation proceeds. Volume 35. has been expanded to 3D [36. For a system of N elements or blocks. no penetration) are satisfied. At the end of each iteration. and ro is the rigid block rotation.. viscosity. there are a total of 6 degrees of freedom or unknowns.37]. If the component of the force normal to the contact is tensile. the spring force is calculated. DT = (uo. Minimization of the potential energy of the system of blocks. no contact check is performed. Tension between two elements can be modeled by applying a lock in the direction where tension is permitted. Full derivation of the equations are provided in . The matrix D represents the unknowns for each element.g. Hence. Coulomb). April 2010 The Arabian Journal for Science and Engineering. ro. In the figure. springs both in the normal and parallel directions are placed to prevent any relative movement at the contact.39]. if smaller than the maximum allowed. an extensive review can be found in ). an iteration process is applied where locks are added or removed as appropriate until all kinematic constrains (e. which was originally developed for 2D problems [32. vo. displacement constraints at the element contacts. body forces. to final failure. Fs > µFn. The contact identification process starts after definition of the elements where some threshold distance is established such that only elements within the threshold distance are checked for contact.. v). εxx. If interpenetration between two elements is detected. the normal spring is removed. The figure shows the different stages of the failure. εyy γxy) and T are the appropriate coefficients from (8). The limitation that the original DDA had that the blocks could not break has been overcome by new developments in modeling.e. Equations (8) are expressed in matrix notation U =TD (9) where U = (u. the DDA has been extended to fragmentation and fracture propagation problems [38. yo. movements between blocks can be avoided. following FEM convention. with laboratory and field measurements (e.. thus. no tension and no penetration between blocks are allowed. the total number of unknowns is 6N. a critical tensile threshold is reached) the elements can separate.35]. respectively. inertia forces. The kinematics of the block system are incorporated into the equations of equilibrium (10) by adding very stiff springs between appropriate elements to lock the movement in the corresponding direction.g.g. Validation of the Displacement Discontinuity Analysis has been done extensively by comparing predictions from the method with analytical solutions. εyy and γxy are the axial strains and the shear strains in the x and y axis. To impose the kinematics of the problem requires addressing two issues: (1) determine contacts between blocks. Number 1B 37 . The procedure of solving the equilibrium equations.. about point xo. The sub-matrices Kij are obtained by minimizing the potential energy associated with strain energy.
Number 1B April 2010 . (1998). 5. Tensile and shear cracks between particles occur when the tensile or shear strength of the contact is reached. (c) t=0. which in turn are used to determine the magnitude of the forces and moments acting between particles. The solution of a problem with static or dynamic loading is done incrementally with very small time steps (for static loading. Adapted from Jing. Newton’s second law of motion is solved through a central finite difference algorithm to determine the displacements and velocities of each particle due to the forces acting on the particle. The updated loads 38 The Arabian Journal for Science and Engineering. Volume 35. Equations (3) and (4) are used to determine the motions of any particle. displacements and velocities of each grain are computed using Equation (3) with the magnitude of the forces equal to those at the end of the previous step. the relative motions between particles in contact is obtained.Antonio Bobet Figure 9.002s. (a) initial geometry. As with the DEM. (d) t=0.005 s. time is an auxiliary variable related to the load increment during each step). and (f) t=0. see Figure 10(b). t=0.003 s. (e) t=0. The forces arise from the weight of the particle and from the contact forces between particles. 8×5m tunnel with a vertical load applied at the surface. The procedure follows that of the Distinct Element Method. see Figure 10(a). Example application of DDA Method. The particles interact with each other through their contacts such that deformation is produced at the particle contacts or by relative displacements between particles.001 s. (b) at time. The key idea of the method is that the geomaterial can be approximated by an agglomerate of cemented grains.004 s. The grains or particles are assumed rigid with circular or spherical shape with a non-uniform distribution. DISCONTINUUM: BONDED PARTICLE METHOD The Bonded Particle Method  originates from the application of the DEM to a discontinuous medium modeled as discs in two dimensions or spheres in three dimensions. From absolute displacements.
Bonded Particle Method Discretization where Fi is the inter-particle force between particle A and particle B (Figure 10(b)). Figure 10(b). with components Fin and Fis in the directions normal and parallel. k s . and polar moment of inertia of the bond between the two particles. Volume 35. A. Fi and Mi are the force and moment carried by the bond between the two particles. During the process. contact between particles is reviewed and updated as new contacts may be formed or old ones are destroyed. as bonds between particles break. respectively. Inter-particle forces and moments are obtained based on the relative motions between particles and on the properties of the particles and bond. Number 1B 39 . The magnitude of the forces and moments. is given by Fi = Fn n + Fs s i i n n + Fs s Fi = F i i n n + Ms s Mi = M i i (11) Figure 10. to the contact between the two particles.Antonio Bobet are then used to compute motions for the next time increment. and are given by: April 2010 The Arabian Journal for Science and Engineering. ∆Un and ∆Us are the incremental normal and shear displacements between particles. and ∆θn and ∆θs are the incremental rotational angles also in the normal and shear directions. moment of inertia. The magnitude of the loads is given by : ∆F n = i kA kB n n k A +k B n n ∆U n kA kB ∆F s = . k n and k s are the normal and shear stiffnesses of particles A and B. The process is repeated until the complete solution of the problem is obtained. I and J are the area.s s ∆Us i A B k s +ks ∆F n = k n A ∆U n (12) i ∆F s = -ks A ∆Us i ∆M n = -ks J ∆θ n ∆M s = -k n I ∆θ s A A B B k n . and k n and k s are the normal and shear stiffness of the bond between particles.
τ c . the following microproperties are needed for the model: kn. bond breaks and it is removed from the model. The list of applications of the method is not exhaustive. constitutive model. processes the information. which are associated with the grains. σc . thus. The disadvantages are that ANN systems operate as “black boxes” in that there is no possibility of assessing how they work internally. The maximum tensile and shear stresses acting on the bond are calculated as σ max = . and their performance may not be easily predicted . A uniform particle size distribution was used to model the rock. and sends the results to other neurons. as shown in Figure 10(b). Figure 11 illustrates the use of the model  to determine the damage zone around a circular opening.52–56].Antonio Bobet ⎧2 R ⎪ A =⎨ ⎪π R 2 ⎩ in 2D in 3D ⎧2 3 in 2D ⎪3 R ⎪ I =⎨ ⎪ 1 π R 4 in 3D ⎪4 ⎩ in 2D ⎧n/a ⎪ J = ⎨1 4 in 3D ⎪2 π R ⎩ (13) R is the bond radius. there is a gap between the two particles and the normal and shear forces are set to zero. Number 1B April 2010 .48–50]. Coulomb with Fis ≤ µ Fin. The model has been applied to investigate the strength of soils and rock materials [19. and R . ANNs are composed of a number of interconnected units called neurons. blasting and dynamic analysis [58–60]. the two particles overlap and. . it has been already used for a wide range of applications within geotechnical engineering. damage to rock mass during tunnel excavation and tunnel support [19. there are normal and shear forces between the particles. The method has been the focus of recent conferences where a large number of cases and applications has been presented. k n . 6. slope stability . What distinguishes one method from the other is how mathematically this is accomplished. e.5 MPa confinement. The methods discussed so far all reach a solution addressing the mechanics of the problem where equilibrium. are fault-tolerant and can reach a solution with ill-defined or imprecise information.57]. and boundary conditions are rigorously satisfied. if it is positive.i + A τ max = Fs i A + Fn Ms R i I Mn R i J (14) When the maximum tensile or shear stress reaches the tensile strength of the bond. ks . The characteristics of ANNs are that the information is stored over the entire network. Thus. which depend on the bond. or shear strength.g. strain compatibility. and it is intended to provide a measure of the wide range of fields where the method is used. OTHER METHODS: ARTIFICIAL NEURAL NETWORK Artificial Neural Networks (ANN) are based on a paradigm completely different than the other numerical methods visited. even in fields beyond civil engineering. ks. ANNs are based on biological models such as the human brain and rely on information processing techniques based on establishing associations between parameters. µ is the coefficient of friction between particles). in the form of tensile and shear cracks. respectively. Volume 35. The model reproduces the experimental results conducted on Berea sandstone where an opening of 14 mm diameter was placed into a prismatic block which was loaded in plane strain with 7. and RA and RB are the radius of particles A and B. are massively parallel processing systems. training may be difficult or impossible.g. their design guidelines and operation are somewhat arbitrary. If the relative displacement between two particles is negative.2 mm. fracture mechanics [19. and the behavior of granular materials and powders [61–63]. and can learn and adapt. Figure 11(b) shows the final stage of failure of the opening with significant cracks and notches between grains. Each neuron receives information.. the The shear force Fis in (12) is limited by the constitutive law used for inter-particle friction (e. As with the human brain. They may be perceived as highly sophisticated 40 The Arabian Journal for Science and Engineering. Even though the Bonded Particle Model is relatively new. and µ. similar to the actual size of the Berea sandstone grains. with average particle size 0. σ c and τ c .
⎟ ⎠ ( ) j ∈ Lk (15) where f is the activation function. θj is the bias or threshold value associated with neuron j. If the state of neuron j is denoted by ij ⎛ ⎜ i j =f⎜ w hj ih + θ j ⎜ ⎝ h ∈L k −1 ∑ ( ) ⎞ ⎟ ⎟ = f oj . There can be any number of hidden layers and each layer can have any number of units. The process in the network works as follows: an array of input values is defined as the state of the neurons in the input layer. Number 1B . and m output units.66]. is defined within the code. and is based on a series of two or more layers of neurons (Figure 12(a)). The state of the neurons in the output layer constitutes the output of the system. following the protocol defined in Equation (15). The units or neurons in each layer are forward connected only to the units or neurons in the next layer. neuron j in layer Lk receives input from the neurons in layer Lk-1. Example of BMP Model. each corresponding to an input parameter..Antonio Bobet curve fitting techniques but they have proven to provide reasonable solutions to imprecisely formulated problems or to phenomena only described through observations . is passed forward to its connected neuron in the next layer and modified by a connection weight and a bias or threshold value. which requires training of the ANN. the activation function. Volume 35. The resulting value is further modified in the receiving neuron by a function called the activation or transformation function (Figure 12(b)). As shown in Figure 12(a). hidden layer.5 MPa. and oj is the argument of the function. The weights and biases are not known. however. The layers between the input and output are called hidden layers. The information stored in each neuron. the input consists of n units. Several 41 April 2010 The Arabian Journal for Science and Engineering. Circular tunnel subjected to biaxial compression with σ3 = 7. but it was not until the 1980s when the introduction of new architectures and learning processes made ANNs useful and practical tools. the learning or training scheme. ANNs are connected to the exterior by the input and output layers only. There is no connection between neurons in the same layer. For example. There are several types of Artificial Neural Networks depending on the characteristics of each neuron. often called the state of the neuron. Thus. The first layer receives the input applied to the network and the last layer contains the output. each corresponding to a requested result parameter. the state of the neurons in the second layer is transmitted to the third layer where new calculations are performed to obtain the state of the neurons in this layer.g. network topology. (2002). The process is repeated until the output layer is reached. The first generation of artificial Neural Networks started in the 1940s to 1960s. Figure 11. e. and network function. whj is the weight associated with the connection between neuron h in layer Lk-1 and neuron j in layer Lk (note that wjh does not exist since there is no connection back from neuron h to neuron j). The Feedforward Network is still the preferred type in geoengineering. From Fakhimi et al. [65. These values are transmitted to the second.
to obtain the values of weights and biases) is done by comparing the output provided by the ANN with actual results.Antonio Bobet functions are possible (e.g. Artificial Neural Network The sigmoidal function is part of the family of squashing functions which constrain the output to values in the range 0 to 1. etc. which is necessary for the training of the ANN. It is a continuous function and its first derivative exists. Training of the network (i.. associated with a given input. Volume 35. multiplicative. Number 1B April 2010 . The function used the most is the sigmoidal function.. The strategy normally used is to minimize the difference between actual and predicted results using the error norm E = m ∈L N ∑ [ t m -f(om )]2 (17) 42 The Arabian Journal for Science and Engineering. which has an expression f oj ( )= 1 1+e -o j (16) Figure 12.e.). tm. linear.
ANNs have also been coupled with FEM. the following needs to be determined: geometry of the problem. In any analysis. PFC). given results or measurements. The input and output neurons. EXAMINE2D and EXAMINE3D (Rockscience). the user needs to decide what are the variables that may affect the results and what are the results needed.. The choice between one code or another. April 2010 The Arabian Journal for Science and Engineering..Antonio Bobet There are different strategies to minimize the error E in (17) by changing the values of the weights and the biases. 84–86]. Distinct Element Method: EDEM (DEM Solutions). slope stability [76. PHASE2 (Rockscience). where the derivatives of the error function E with respect to the weights or biases are set to zero. All codes are based on the principles of mechanics and they rigorously solve (in the context of numerical solutions) equilibrium equations. Inc. and construction process. There are no rules to design ANNs. PLAXIS. and number of passes (epochs) for training. depth. where the Finite Element Method is used to solve the mechanics of the problem or to produce the data for training the ANN. PHASE. boundary conditions. etc. deep excavation deformations . strain compatibility. ANNs are being used in many fields of geoengineering.77]. results are obtained (e. and Bonded Particle Method: PFC2D and PFC3D (ITASCA Consulting Group. Karlson and Sorensen. displacements). ANNs have been applied to obtain soil and rock properties [68–73] including soil liquefaction [74. A number of strategies can be tested with a different number of hidden layers. or on a backward analysis where. Finite Element Method: ABAQUS (Hibbit. PENTAGON-2D and -3D (Emerald Soft). Once the ANN is trained and selected. are defined by the user. Number 1B 43 . and yet it is accurate enough that the results sufficiently approximate the performance of the design. All codes have a very steep learning curve and it may take significant time and effort for a company to train engineers in any one particular code. Computer Software and Services (CSS)). The number of hidden layers and the number of neurons per layer is problem-dependent. and the constitutive material model. The codes listed can be divided into Continuum (FLAC. visco-elastic. UDEC. The methods can be used both in a forward analysis where.g. they have been successful in giving accurate predictions to problems that cannot be solved following the mechanics approach because some of the inputs or conditions needed are not well defined or the input data may be not completely reliable. Thus. Increasing the number of neurons and/or hidden layers does not necessarily result in better predictions. UDEC. 7. Inc. DISCUSSION Numerical methods are tools that the engineer has to evaluate qualitatively and quantitatively the effects of geology on the design and the consequences of the design on geology. The strategy often followed consists of dividing the available data in two sets: one for training and the other for validation. Boundary Element Method: BEFE (coupled BEM-FEM. In any case. mining and tunneling support [79–83]. For example. it can be used for predictive purposes.and post-processing capabilities. The goal is to create a model that is simple enough such that it can be implemented and interpreted within a reasonable amount of time. All numerical models visited in this chapter are capable of providing reasonable results when sound engineering judgment is employed with their use. In fact. ground properties or ground behavior are approximated. EXAMINE. i. A word of caution needs to be added for Artificial Neural Networks since their use should be confined within the range of the database employed for their training. The largest portion of time spent in modeling is during pre-processing or discretization and post-processing or results analysis. the actual construction process cannot be predicted. This is so because in many cases the geology and material behavior are not fully known. Without exception. and tunneling [67. Inc. or the ANN is used to obtain input parameters for the FEM from back-calculation or to make predictions based on input data from the FEM .). 3DEC. PLAXIS (Plaxis BV).e. etc. Each trained ANN is then tested against the validation data.). overfitting the ANN is a real danger which may induce erroneous results. PENTAGON. in terms of numbers and characteristics.). There is no guarantee.75]. It is very important to realize that the ANN should not be used to make predictions outside the range of cases within which it has been trained. or the numerical model is necessarily applied to a limited volume of the entire domain. The most common strategy is the backpropagation algorithm or delta rule [65. the error norm in (17) is minimized. assumptions and decisions need to be made. These are almost exclusively commercial codes.66]. plastic. 3DEC (ITASCA Consulting Group. extent.). stresses. given geometry and properties. BEFE) and Discontinuum (EDEM. It is perhaps for this reason that the most used numerical methods in practice are those that include user-friendly pre. Despite the shortcomings of the ANNs.. including the geologic geometry in terms of layers. actual material behavior such as elastic.. that the process described will result in at least one of the ANNs providing satisfactory results. Inc. Volume 35. Thus there is a tendency to keep the expertise within a very reduced number of numerical codes. ABAQUS. within the realm of continuum or discontinuum. different number of neurons per hidden layer. is often based on personal or company preferences. appropriate boundary conditions. The following is a list of the codes most referenced in the literature: Finite Difference Method: FLAC and FLAC3D (ITASCA Consulting Group. selecting the ANN with the smallest differences . all the details and complexities of the problem cannot be introduced into the numerical model. however.
Pande.. N. pp. International Journal of Rock Mechanics and Mining Sciences. England: Allen and Unwin. A Scheme to Detect and Represent Contacts in a System Composed of Many Polyhedral Blocks”. Brown. Cundall. 9(1992). 40(2003). Jing. Volume 35. 1(1971). England: Allen & Unwin. 41(2004). Otherwise. 1329–1364. Oxford. 1982. R. G. Venturini. REFERENCES    L. C. Strack. USA: McGraw Hill. 107–116. P. P. B. Beer and J. and L. pp. pp. This issue is still under debate. pp. there is a large experience-based on continuum models successfully used in rock masses. paper No. International Journal of Rock Mechanics and Mining Sciences. ed. S. pp. Jing and J. Lemos. 1990. Englewood Cliffs. in Proceedings of the 2nd ASCE Conference on Electronic Computation. Brebbia and S. Analytical and Computational Methods in Engineering Rock Mechanics. Cundall. 25(3)(1998).                 44 The Arabian Journal for Science and Engineering. France.129–163. P. Crouch and A. but on the other hand there is mounting evidence that in discontinuous media the stress field obtained with a continuous model does not compare well with the stress jumps across discontinuities predicted by discontinuous models [15. 283–353. and J. Boundary Element Methods in Solid Mechanics. pp. “Distinct Element Models of Rock and Soil Structure”. Cundall. V. 1970. 1983. Nancy. M. 409–427. For rocks. Bathe. G. 345–378. Engineering Computations. A. in Lecture Notes in Engineering. A. USA: Wiley. “Numerical Methods in Rock Mechanics”. on the one hand. “Formulation of a Three-Dimensional Distinct Element Model – Part I.. S. Geotechnique. R. 1992. P. International Journal of Rock Mechanics and Mining Sciences. International Journal of Rock Mechanics and Mining Sciences. Cundall and O. Potyondy and P. Number 1B April 2010 . 164–204. 117–125. A. USA: PrenticeHall Inc. pp.T. England: John Wiley and Sons. C. K. Southwell. “The Finite Element in Plane Stress Analysis”. A. P. C. If very few discontinuities are present in the medium. Timoshenko and J. Theory of Elasticity. A. R. New York. Advances and Outstanding Issues in Numerical Modeling for Rock Mechanics and Rock Engineering”. the size of the blocks determined by the discontinuities is much smaller than the characteristic size of the geostructure) a pseudo-continuum model can still be applied. West Sussex. N. G. Hart.Antonio Bobet For soils it is often assumed that a continuum approach is appropriate. F. Orszag. pp. W. Brebbia. 101–113. Mechanical Calculations for Motion and Interaction of a System Composed of Many Polyhedral Blocks”. D. London. H. in Proceedings of the Symposium of the International Society of Rock Mechanics. “A Review of Techniques. R. 1987. NY. Cundall and R. Wrobel. J. II-8. Watson. A. Beer. P. E. Ed. 47–65. Germany: Springer-Verlag. a continuum model can still be efficient. C. A. “Boundary Element Method in Geomechanics”. A. pp. London. Germany: Springer-Verlag. Cundall. eds. in Analytical and Computational Methods in Engineering Rock Mechanics. “A Computer Model for Simulating Progressive Large Scale Movements in Blocky Rock Systems”. Williams. Starfield. 1946..Y. G. a discontinuum model seems more reasonable. L. Boundary Element Techniques. A. L. PA. Brady. London. “A Bonded-Particle Model for Rock”. A. 29(1)(1979). Pittsburgh. New York. Introduction to Finite and Boundary Element Methods for Engineers. England: Clarendon Press.. Berlin. New Jersey. Numerical Methods in Rock Mechanics. T. there are no guidelines to decide when a continuum or a discontinuum model should be used. “A Discrete Numerical Model for Granular Assemblies”. with a large number of discontinuities (e.g. N. Cundall. 1983. P. Finite Element Procedures in Engineering Analysis. Hudson. O. J. “Formulation of a Three-Dimensional Distinct Element Model – Part II. Telles. “Boundary Element and Linked Methods for Underground Excavation”. and J. International Journal of Rock Mechanics and Mining Sciences. pp. 1984.88]. S. Ltd. Brown. W. 39(2002). L. “Numerical Modelling of Discontinua”. Relaxation Methods in Theoretical Physics. O. 25(3)(1988). however.. Goodier. Clough. E. A. 1987. England: Allen & Unwin. Hart. D. 1960. D.
M. “Two Dimensional Discontinuous Deformation Analysis”. Zhao. Heasley Editors. Morris. G. Dowding. “A Coupled Finite Element-Rigid Block Method for Transient Analysis of Rock Caverns”. “Three-Dimensional Discrete Element Method for Granular Materials”. Engineering Computations. Ghaboussi and R. 2003. C. “A Study of UDEC Modelling for Blast Wave Propagation in Jointed Rock Masses”. Doolin. H. 145–155. 7(1983). Hocking. C. “Insights Into Ground Shock in Jointed Rocks and the Response of Structures Therein”. 44(5)(2007). 95–116. Topics in Engineering. S. J. Dowding. 647–676. Stead. H. Number 1B 45 . 35(1998). Morris. 33(1–12)(2006). 14(1990). Golden. I. (1984). H. International Journal of Rock Mechanics and Mining Sciences. M. Goodman. International Journal of Rock Mechanics and Mining Sciences. B. J. in Proceedings of the 25th U. International Journal of Rock Mechanics and Mining Sciences. 38(2001). P. H. L. N. International Journal of Numerical and Analytical Methods in Geomechanics. Review of Rock Joint Models. D. Washington DC. MacLaughlin. Shi and R. pp. O’Connor and C.Antonio Bobet   G. 9(1992). Shi. Paper 06-1078. 243–252. G. Dwyer. 11. C. H. 269–277. “The Validity of Dynamic Block Displacement Prediction Using DDA”. . 2003. Shi. Shi and R. Yeung. Boston. T. G. P. J. International Journal of Rock Mechanics and Mining Sciences. Block. P. Vol. “Three Dimensional Discontinuous Deformation Analysis”. M. Jiang and M. Shi. Block. B. G. Eberhardt. A. pp. Rock Mechanics Symposium. pp. C. and J. Proceedings of the 38th U. E. 1421– 1428. 93–99. Morris and G. USA. A. K. Y. Yen. 37(2)(2004). UK: WIT Press. pp. Elsworth. International Journal of Numerical and Analytical Methods in Geomechanics. J. pp. Koo and J. pp. Heuze and J. Engineering Computations. “A Model of Point-to-Face Contact for Three-Dimensional Discontinuous Deformation Analysis”. E. P. Chern.S. “Hybrid Discrete Element Code for Simulation of Mining-Induced Strata Movements”. M. Feintuch. G. pp. Chen. 753–762. Rock Mechanics and Rock Engineering. 41(2004). 463–473. Jung. Barbosa. Goodman. H. pp. M. UCRL-ID-153650. Hatzor and A. (2006). 38(2001). Symposium on Rock Mechanics. G. Morris. 33(7)(1996). Block System Modeling by Discontinuous Deformation Analysis. USA: Balkema: Rotterdam (2001). Rock Mechanics in the National Interest. pp. T. P. International Journal of Rock Mechanics and Mining Sciences. Sitar. I. D. Q. pp. and J. CO. S. International Journal of Impact Engineering. S. J. International Journal of Rock Mechanics and Mining Science and Geomechanics Abstracts. Engineering Computations. 683–693. “Numerical Analysis of Initiation and Progressive Failure in Natural Rock Slopes – The 1991 Randa Rockslide”. pp. 235–242. P. eds. Preece. 541–556.                     April 2010 The Arabian Journal for Science and Engineering. J. “Discontinuous Deformation Analysis”. Engineering Computations. E. J. C. Lin. and K. pp. pp. D. G. Tinucci. Y. 41st U. Volume 35. F. H. 451–472. Rock Mechanics Symposium. S. Belytschko. and H. “Modification of the DDA Method for Rigid Block Problems”. 1993. http://www-r. “The Discrete Element Method for Analysis of Fragmentation of Discontinua”. International Journal of Rock Mechanics and Mining Sciences. E. 9(1992). “Extensions of Discontinuous Deformation Analysis for Jointed Rock Masses”. Bonner. Connor. 599–606. H. “Investigation of Slope-Stability Kinematics Using Discontinuous Deformation Analysis”. Amadei. 157–168. and T. 10 pages. 671–694.pdf. R. M. Rubin.S. pp. Brebbia and J. J. and J. International Journal for Numerical and Analytical Methods in Geomechanics. and M. 9(1992). Taylor and D.H. 117–127. Mohammadi. pp. 69–87. pp. Discontinuum Mechanics: Using Finite and Discrete Elements. Computational Mechanics Publications. 35(1)(1998). B. Abbot. pp.gov/tid/lof/documents/pdf/244645. “Discontinuous Deformation Analysis: A New Numerical Model for the Statics and Dynamics of Deformable Block Structures”. “Simulations of Fracture and Fragmentation of Geologic Materials Using Combined FEM/DEM Analysis”. 9(1985). “Simulations of Underground Structures Subjected to Dynamic Loading Using Combined FEM/DEM/SPH Analysis”. eds. 9(1992). Coggan. Southhampton.S. “Simulation of Blasting Induced Rock Motion Using Spherical Element Models”. pp.llnl. American Rock Mechanics Association. Lawrence Livermore National Laboratory.
A. pp. Holt. “Review of Validation of the Discontinuous Deformation Analysis (DDA) Method”. ed. pp. F. Engineering Computations. pp. McPherson. pp. ed. “Dynamic Modeling of Induced Seismicity”. D. Konietzky. Numerical Modeling of Discontinua. Development and Application of Discontinuous Modelling for Rock Engineering. Wang. C. D. P. and N. Holder. M. J. pp. L. 40(2003). Jing. Konietzky. “DEM Study of Wave Propagation in Weak Sandstone”. International Journal of Rock Mechanics and Mining Sciences. pp. 117. “Formulation of Discontinuous Deformation Analysis (DDA) – An Implicit Discrete Element Model for Block Systems”. Kleinschrodt. and J. M. ASCE. J. Y. “Thin Rock Support Liners Modeled with Particle Flow Code”. Comacho. International Journal of Numerical and Analytical Methods in Geomechanics. Yeung. M. “Evaluation of Schemes to Improve the Efficiency of a Complete Model of Blasting and Rock Fracture”. Netherlands: Balkema. P. Rock Mechanics and Rock Engineering. and G. Netherlands: Balkema. Pilliteri. Geotechnical Special Publication No. Boutt and B. Konietzky. Volume 35. F. Rauch. “Comparison Between Controlled Laboratory Experiments and Discrete Particle Simulations of the Mechanical Behavior of Rock”. 117. 1365–1376. pp. Li and R. 507–515. Lilly. D. Wang. 21(2/3/4)(2004). Berger. Proceedings of the 6th International Conference on Analysis of Discontinuous Deformation. J. Tannant and C. M. Sun. “Simulation of Failure Around a Circular Opening in Rock”. D. “Validation of Block Theory and Three-Dimensional Discontinuous Deformation Analysis as Wedge Stability Analysis Methods”. Lu. 2003. Kjølaas. M. Netherlands: Balkema. pp. A. P. A. J. A. pp. Doolin. Carvalho. ASCE. pp. L. Fakhimi. 131(10)(2003). 37–43. L. Wanne. 40(2003).19–27. 1234–1242. eds. L. 86–92. H. F. pp. pp. R. pp. F. R. F. Jiang. “Modeling Rock Failure Around a Circular Opening”. M. J. T. ed. International Journal for Numerical and Analytical Methods in Geomechanics. Konietzky. E. 323–328. Experimental and DDA Results”. Sitar. Geotechnical Special Publication No. 31(5)(2007). M. A. ed. 38(2)(2005). H. pp. A. and P. A. Konietzky. Journal of Geotechnical and Geoenvironmental Engineering. Wang. Q. “The Role of Particle Packing in Modeling Rock Mechanical Behavior Using Discrete Elements”. 2002. Tannant. B. Guest. J. D. Numerical Modeling in Micromechanics via Particle Methods. Tannant and C. MacLaughlin and D. M. D. ASCE.Antonio Bobet  M. Numerical Modeling in Micromechanics via Particle Methods. 371–381. “A Decade of DDA Validation”. 30(2006). M. Hazzard and R. 153–167. Bertrand. Fakhimi and J. A. Olson. ASCE. Ruest. ASCE. 985–995. 346– 352. pp. 265–275. Geotechnical Special Publication No. Narayanasamy. Konietzky. Maynar and L. (2002). 318–342. ed. 42(2005). and G. “Numerical Analysis of the Stability of Heavily Jointed Rock Slopes Using PFC2D”. M. H. Ishida. pp. “Development of Discrete Particle Modeling Towards a Numerical Laboratory”. Kleier and H. Netherlands: Balkema. pp. Discrete Element Methods: Numerical Modeling of Discontinua. A. 117. 2002. Numerical Modeling in Micromechanics via Particle Methods. Reston. International Journal of Rock Mechanics and Mining Sciences. Wu. pp. Discrete Element Methods: Numerical Modeling of Discontinua. and B. Labuz. 13–31. “Dynamic Displacement of a Block on an Inclined Plane: Analytical. J. 107–115. and N. 271–305. H. “Discontinuous Mechanical Modeling of Granular Solids by Means of PFC and LS-Dyna”. pp. Sønstebø. International Journal of Rock Mechanics and Mining Sciences. 41(2004). 241–249. pp. Hatzor. 2002. (2002). A. “PFC3D Simulation Procedure for Compressive Strength Testing of Anisotropic Hard Rock”. H. D. Balkema. H. Engineering Geology. K. Cook and R. and E. H. 39(2002). Numerical Modeling in Micromechanics via Particle Methods. M. H. T. 117.                     46 The Arabian Journal for Science and Engineering. “Thin Tunnel Liners Modeled with Particle Flow Code”. D. MacLaughlin and E. “Modeling of Cyclic Fatigue Under Tension with PFC”. Young. Geotechnical Special Publication No. Jensen. R. A. Discrete Element Methods: Numerical Modeling of Discontinua. 49(1998). “Discrete Numerical Model for Analysis of Earth Pressure Balance Tunnel Excavation”. 415–424. International Journal of Rock Mechanics and Mining Sciences. Rodríguez. 649–666. Number 1B April 2010 . pp. Netherlands: Balkema. The Netherlands: A. International Journal of Rock Mechanics and Mining Sciences. “Applying Discontinuous Deformation Analysis to Assess the Constrained Area of the Unstable Chiufenerh-shan Landslide Slope”. USA. 2002. 335–339. Labuz. G. R. Li. M. H. Holt. 37–43. (2002). te Kamp. VA. A. Tsesarsky. ed. Numerical Modeling in Micromechanics via Particle Methods. Discrete Element Methods. M. 2002. Chitombo. 2000. Cundall.
Altun. Tunnelling and Underground Space Technology. P. 21(2006). R.-G. G. 207–222. H. Alvarez Grima. J. Deb. 24(2006). pp. 1021–1037. “Application of Neural Networks for the Prediction of the Unconfined Compressive Strength (UCS) from Equotip Hardness”. Y.. Bai. Geotechnical and Geological Engineering. International Journal of Rock Mechanics and Mining Sciences. 36(1999). ASCE Journal of Geotechnical and Geoenvironmental Engineering. A. G. Ariaratnam. and R. “The Use of PFC2D to Simulate Milling”. Yue. and H. Sain. Tezekici. Konietzky.                  April 2010 The Arabian Journal for Science and Engineering. Artificial Neural Networks. pp. Lueke and S. Nefeslioglu. “A Study of Slope Stability Prediction Using Neural Networks”. pp. Fener. and M. Sakellariou and M. Q. Singh. pp. 29–39. pp. Inc. 2002. H. Germany. Journal of Geotechnical and Geoenvironmental Engineering. A. A. and R. Singh. pp. “Seismic Liquefaction Potential Assessed by Neural Networks”. 43(2006). Wang. “Prediction of Strength Properties of Some Schistose Rocks from Petrographic Properties Using Artificial Neural Networks”. 1996. Tunnelling and Underground Space Technology. “Use of Artificial Neural Networks in the Prediction of Liquefaction Resistance of Sands”. 133–150. 259–268. An. H. S. and T.-T. 43(2006). H.Antonio Bobet       M.Y. 24(2006). N. Kumar. 20(4)(1997). Young-Su and K. “A Neural Network Model for Real-Time Roof Pressure Prediction in Coal Mines”. Numerical Modeling in Micromechanics via Particle Methods. T. “The Artificial Neural Network as a Tool for Assessing Geotechnical Properties”. H. pp. T. A. J. 149–168. 20(2002). Yang and M. “Hybrid Intelligent Method Optimization of a Soft Rock Replacement Scheme for a Large Cavern Excavated in Alternate Hard and Soft Rock Strata”. 1467–1480. Lee. Deng. F. C. “Displacement Back Analysis for a Steep Slope at the Three Gorges Project Site”. pp. C. 38(2001). “Estimation of Rock Modulus: For Intact Rocks with an Artificial Neural Network and for Rock Masses with a New Empirical Equation”. C. pp. Gupta. China”. Konietzky. 120(9)(1994). Zhang.. Rosha. Meulenkamp and M. 23(2005). Neural Networks and Reliability: A Case Study of the Feng Huangshan Copper Mine. X. pp. H. Goh. 389–395. A. 106–117. S. 157–164. Gokceoglu. Geotechnical and Geological Engineering. 400–409. 269–284. R. D. 40(2003). Yang and Q. 585–599. Bwalya. X.-T. S. USA: The McGraw-Hill Companies. 20(2005). Feng and H. Gelsenkirchen. K. V. “Artificial Neural Networks for Predicting the Maximum Surface Settlement Caused by EPB Shield Tunneling”. pp. pp. Singh. 1502–1504. The Netherlands: Balkema. N. D. Sonmez. G. 224–235. M. J. P. Suwansawat and H. “A Hierarchical Analysis for Rock Engineering Using Artificial Neural Networks”. and J. Y. ASCE Journal of Geotechnical Engineering. Singh. “Modular Neural Networks for Predicting Settlements During Tunneling”. Rock Mechanics and Rock Engineering. Schalkoff. Volume 35. 21(2006). Neural Network Models: An Analysis. Ortigao. B. pp. London. 41(2004). International Journal of Rock Mechanics and Mining Sciences. C. H. pp. De Wilde. and J. Einstein. Proceedings of the 1st International PFC Symposium. “Sawability Prediction of Carbonate Rocks from Shear Strength Parameters Using Artificial Neural Networks”. 73–77. R. Kayabasi. D. Shi. 1997. “Estimating Wall Deflections in Deep Excavations Using Bayesian Neural Networks”. 647–653. pp. 299–312. Number 1B 47 . J. and A. Geotechnical and Geological Engineering. “Pillar Design by Combining Finite Element Methods. pp. Rosenbaum. 33(6)(1996). “Numerical Modeling in Micromechanics via Particle Methods”. Zhu. International Journal of Rock Mechanics and Mining Sciences. Kahraman. “A Comparative Analysis of Cognitive Systems for the Prediction of Drillability of Rocks and Wear Factor”. Deng and C. S. Netherlands: Balkema. Byung-Tak. International Journal of Rock Mechanics and Mining Sciences. International Journal of Rock Mechanics and Mining Sciences. T. pp. 419–445. N. 132(11)(2006). Yao. Tunnelling and Underground Space Technology. England: Springer-Verlag. New York. S. Y. K. Z. pp. L. ed. Ferentinou. Moys. Feng. H. J. Tham. M. 655–667. 2002. Goh. “Numerical Characterization of Surface Heave Associated with Horizontal Directional Drilling”. Chua and T. F. “Forecasting Shield Pressures at a Longwall Face Using Artificial Neural Networks”. Geotechnical and Geological Engineering. pp. 38(2001).-J. International Journal of Rock Mechanics and Mining Sciences. J. H. S. International Journal of Rock Mechanics and Mining Sciences. International Journal of Rock Mechanics and Mining Sciences. 124(5)(1998). and M. pp.
“Chapter 9: Soft Computing-Based Parameter Identification as the Basis for Prognoses of the Structural Behavior of Tunnels”. M. “Rock Mass Characterization and Modelling Aspects of Mining and Civil Engineering”. Volume 35. Cartagena. Tunnelling and Underground Space Technology. G. pp. R. 21(2006). 201–223. “Prediction of Tunneling-Induced Ground Movement with the Multi-Layer Perceptron”. B. and H. 2006. 45–75. Neaupane and N. Adhikari. A. Lackner. Colombia. 2003. Numerical Simulation in Tunnelling. XI Congreso Colombiano de Geotecnia . Barton. pp.VI Congreso Suramericano de Mecánica de Rocas. Beer .  48 The Arabian Journal for Science and Engineering. Wien. 8-13 October. 151–159. Austria:Springer-Verlag. N. Pichler. R.Antonio Bobet   K. Ed. pp. Mang. Number 1B April 2010 .