REAND 7. 1. LoNawomnff CUNDALL,P. A. & Sreack, O.D. (1979). Géotechnique 28, No. 1, 47-65, 1 clay be cared outy of hydraulic uit. Te casured by piezomete ‘Hissures may be miss ‘Tein the borehole, + it is usually necessay and Hobbs, 1977. i ays Sand, wasreadera igh water presies i >ase ofthe pitas no, ‘ted in the mary sini aad water movement i 's indicating hydraulic uunme of the Building Paper is publsted by (the assistance of the ‘carried out with the lue to Mr J. Herel, scussions durig the {and progressive allure ‘shire Limestone. Water ‘sof aheavy foundation ‘Department of Mining ‘Empingham, Reland, ‘ugh. Institute of Geo- tial photoprametry. site investigation prac- ‘a trench excavation, ‘ofthe Environment, dries, Patstand A discrete numerical model for granular assemblies P, A. CUNDALL* and O. D. L. STRACKt toe distinct element method is a numerical modet capeble of describing the mechanical behaviour of ssemblie of diss and spheres. The method is based onthe useof an explicit numerical scheme in which the interaction of the particles is monitored contact by eontact and the motion of the particles modelled particle by particle. The main featares of the distinct flement mathod are described. The method is vali- {ated by comparing force vector plots obtained from the computer progrim BALL with the corresponding plots obtined from a photoelastic analysis. The photoelastic analysis wed for the comparison is the ‘ne applied to an assembly of discs by De Josselin de Jong and Verrijt (196). The force vector diagrams ‘obtained numerically closely resemble those obtained photoclasically. It is concluded from this eompari- f0a thatthe distinct element method and the program BALL arevalid tools for esearch nto the behaviour of La méthode des éiéments distinct est un modéle numérique capable de décrire le comportement smcanigue de Lassemblage de dsgues et de spires La mithode est basée sur Mutsation d'un systtme rumérique expicte dans Teqvel Tinteraction des particles ext conteblée conlace par contact et le mouvement des particule simulé particle par part- ule. Les caracteritiques principales de la méthode ‘yalldge en comparant les tacts do veceur de force ‘obtenus parle programme sur ordicateur BALL avec Jes tracéscorrespondants obtanus & Taide une Analyse photo-Sastigue. analyse, photodlasique ‘tlisée pour la comparaison est celle appliqués sur lun aserblage de disques par De Josslin de Jong et ‘Verruijt (196). Las diagrammes de vcteur de force obtenus numfrquement sont tts voisins de ceax fobtenus ‘photoslastiquement. Cele comparsison permet de conclure que la methode des éléments distinct et Ie programme BALL soat ds instruments Yalables pour la recherche du comportement des ‘panular assemblies. assemblages granular. INTRODUCTION ‘A granular medium is composed of distinct particles which displace independently from one another and interact only at contact points. The discrete character of the medium results in a complex behaviour under conditions of loading and unloading and to date no satisfactory con- stitutive relationships have been established. Extensive programmes of laboratory tests are necessary to generate new constitutive relations or to validate existing ones Interpretation of tests on real granular media, such as sand, is dificult because the stresses inside the sample cannot be measured and must be estimated from the boundary conditions. New experimental techniques using x-ray photography make measurements of strains possible ‘but no such techniques exist forthe determination of stresses within a sand sample. ‘The uncertainties regarding interior stresses in samples of sand have led to the development of models of granular media that are geometrically simpler than sand and for which the stresses and displacements can either be calculated or measured. These models consist of assemblies of discs or spheres and may be analytical, physical or numerical. ‘An analytical model for cubic arrays of spheres of uniform size was proposed by Deresiewicz ! (1958). The resulting expressions predict a non-linear and hysteretic stress-strain behaviour. Ultimate failure is als¢ accommodated in the formulation. The analytic approsch, however, is restricted in that the aray is cubic, the spheres are uniform in size and the loading sequence is limited. Discussion om this Paper closes 1 Jun, 1979. For further details se inside back cover, * Principal Engineer, Advanced Technology Group, Dames & Moore, London, ¢ Assistant Professor, Department of Civil and Mineral Eoginering, University of Minnesota, USA illSS a P. A. CUNDALL AND 0, D. L. STRACL A testing technique for assemblies of discs that makes possible the direct determination erformed a seis ating model. The atures of the con. esis by numerical sal modelling and "stage of the test, particle sizes, size ‘ped a numerical Xs are calculated amined by inere tees are used for coties of Mindlin A major draw- mber of particles time-consuming lenever a contact ‘particles of any mechanics prob- dasa transient balance. The FY space is such re memory of a hod models the ‘merical repro- Verruijt (1969). + disc tests indi= ‘The numerical vo-dimensionat of the fictitious | sti referred to | puter program tsofa stressed cements of the ‘he medium of DISCRETE NUMERICAL MODEL FOR GRANULAR ASSEMBLIES 9 disturbances originating at the boundaries: a dynamic process. ‘The speed of propagation is a function of the physical properties of the discrete medium. Tin describing the above dynamic behaviour numerically, time steps are taken over which selocities and accelerations are assumed to be constant. The distinct element method is besed upon the idea that the time step chosen may be so small that during a single time step disturb- ances cannot propagate from any disc further than its immediate neighbours. Then, at all times the resultant forces on any disc are determined exclusively by its interaction with the discs with which itis in contact. Itis tis key feature ofthe distinct element method which makes it possible to follow the non-linear interaction of a large number of discs without excessive memory requirements or the need for an iterative procedure. It should be noted that all analyses, presented in this Paper are for dry assemblies, and deal with total contact forces. AA technique that has been devised for accommodating pore pressures will be described in a later publication. The calculation cycle ‘The calculations performed in the distinct element method alternate between the application of Newton’s second law to the discs and a force-displacement law atthe contacts. Newton's second law gives the motion of a particle resulting from the forces acting on it. The force- displacement law is used to find contact forces from displacements. ‘The deformations of the individual particles are small in comparison with the deformation of a granular assembly as a whole. The latter deformation is due primarily to the movements of the particles as rigid bodies. Therefore, precise modelling of particle deformation is not ‘necessary to obtain a good approximation of the mechanical behaviour. In the current version of BALL, the particles are allowed to overlap one another at contact points. This overlapping ‘behaviour takes the place of the deformation of the individual particles. The magnitude of the ‘overlap is related directly to the contact force in the way explained below. It should be noted, however, that these overlaps are small in relation to the particle sizes. ‘To illustrate how forces and displacements are determined during a calculation cycle, the case represented in Fig. 1(a) wll be considered. Two weightless discs, labeled as disc x and disc y, are squashed between a pair of rigid walls. The walls move toward each other at a constant velocity v. Initially, at time ¢ = fo, the walls and discs are touching and no contact forces exist. A time later, the walls have moved inward over distances vAf. In accordance with the assumption thatthe disturbances cannot travel beyond a singledisc during one timestep, both discs are assumed to maintain their initial positions during the time interval from # = fo tot —to4-4t. Overiaps therefore exist at time t, = fo-+ At at contacts A and C (see Fig. 1(6)) and are of magnitude dn = v At Points Aco and Ac (se Fig. 1(6)) are points of the disc and the wal, respectively, lying on the line drawn perpendicular to the wall and through the centre of the dise. The contact A is defined as the point halfway between Avo, and Ac. The relative displacement (4m) at the contact (the overiap) is defined as the displacement of point Acq relative to that of point Aw) ‘cocurring over one time increment. The subscript ¢1 in (4mas)s Tefers to time, “The relative displacements oocurring at contacts A and C at time f, = fo-+4t (see Fi. 1(8)) are used ina force-displacement law for the calculation of contact forees. Inthe present version of the computer program BALL, an incremental force-displacement law of the following form is used arctan shod 0) where k, is the normal stifness and AF, represents the increment in normal force. ‘Defining the postive I direction as pointing from disc x to disc y (see Fig. 1(b) the sums ofP. A. CUNDALL AND 0. D. L TRASK gh = hh Rigle Gaal, Teagly O co @ Eig 1 Two dacs compressed betneen rigid wal (the overlaps are exaggerated (a) f= 20) =H sat At orerepeney Fores Fon and Fin for discs x and y a time t = f+! become, taking Fiyy and Fy, 108 positive in the postive 1 direction, Fon = k4m, Fons ‘These forces are used to find new accelerations. kd -. @ 18 Newton's second law A= Footman i= Foulmy ~ 2. O ‘where # and J, stand for the respective accelerations of discs x and y in the 1 direction and here the subscripts (x) and (y) in the masses ma and mg refer to discs x and y. ‘The accelera- tions found from equation 3) are assumed to be constant over the time interval fom f = fet dt to, = fo 2dr and may be integrated to yield velocities, Gila = CFiodmaldh Gila = Folmgldt {TE lative displacement increments at contacts A, B and C at time fy from fo+2t are found (Anaya = (0— [Fees fm a(10)4t i. e 6 Ar a)a = Fan lma]4t— CF n/m 1d) a) (ng)a=(CFylmy]4t—[-opat. . a, where Aria» Ana and Ano are taken as postive for compression, This yele may be repeated again and agnin: forces corresponding tothe displacements are found using the force displacement law, equation (2), andthe force suns forthe two diss are substituted in Newton's second law, equation (3), to obtain displacements, The above cxample isan illustration of the eycling through a force displacement law and Re ia¥ of motion. In the general case of an assembly of many discs, the force displacement law is applied ateach contact of any dsc and the vectorial sum of these contact forces is deter ey Gao ew accelerations are calculated from Newton's second law. The equations used in ‘caleulation cyte fora general assembly are presented inthe following section, The force-displacement law ‘The force-displacement law wil be presented for the case of two dises in contact, dses x and A DISCRETE NU Fig. 2. Thefor yin Fig. 2.7 yD where the indicated in F j= Gada) direction, wht Ry and Ry section of the Two dises the sum of th If this conditi ‘integration of of point Pus as pointing fr and the unit ‘The relativ ‘The norms X, onto ear and‘waht ily for Oem em Fos 804 Fy tobe @ a e@ the 1 direction and ad y. The accelera- valfeom t, = to+4t rn) fo+2dt are found ) © o) displacements are or the two discs are slacement law and ‘orce-displacement ‘act forces is deter- accomplished for equations used in ion. ontact, dises x and | | | s1 Fig.2. The foree-dlsplacement law yin Fig. . The co-ordinates ofthe dse centres are represented a8 x;= (1, %3) and ¥,= Oy ys) where the indices 1 and 2 refer to the co-ordinates of a Cartesian co-ordinate system as Undated in ig 2. Te component of he wciy weston fick x and = 2), = (ars) and the angular velocities are 6, and Op, taken positive ia counterclockwise “irection, where the dots stand for differentiation with respect to ime. Discs xand y ha Rep and Ry and MAES My and my. Points Pye and Py are defined as the section of the line connecting the dsc centres with the boundaries of dises x and y, respectively. “Two discs are taken to be in contact only ifthe distance D between thir centres is lss than ieif Deke ® If this condition is met, the relative displacement at the contact C (see Fig. 2) is determined by integration ofthe relative velocity. The relative velocity atthe contact is defined as the velocity of point Pe, with respect to Puy. The unit vector e; = (cos a, sin a) (see Fig. 2) is introduced 1s pointing from the centre of dise x to the centre of disc y, ie. gata osmsing oO and the unit vector ¢, is obtained by a clockwise rotation of e, through 90°, i. barnes Lae . 00) The relative velocity of point Puy with reepect to Pi» now may be expressed as X, with X= Gi H)-GeRey Hono Mo AD ‘The normal (i) and tangential (8) components of the relative velocities are the projections of onto e, and &, respectively, Kem GDA OoReot HyRodner= CH «2 and It OnRey OonRedite= i Ht GonRen pod (13)2 PA. CUNDALL AND 0. D. L STRACK ‘summation convention is adopted for the index i; ie. the summation is Performed over the index that occurs twice in each expression (indices in parentheses ae not summed). Integration of the relative velocity component with respect to time gives the con ponents dn and 4s ofthe relative displacement increment Warm (Giese. and eee @ar= ‘These relative displacement increments are usedwith the Tore displacement law to calcu late increments of the normal and shear forces, AF. and AF, AF =k, dn= hf Jedd... 9 BF, by As = kG FD OraRey ty Ry}. (AT) where ky and k represent the normal and shear stffnesss, respectively Finally, at each time step the force increments AF, and AF, are added into the sum of all force increments, F, and F,, determined for previous time steps: and Cov = Fov-at Fy (Foy =(Fn-at dF. - a9 here theindices Wand N-—1 refer to times ty and y_, such that fy—ty- = At, Thesigncom- Nention for the normal and shear forces acting on die x is as indicated in Fig. 3: F, and F, ‘are taken as positive in the directions opposite to e, and ‘4 Coulomb-type friction law is incorporated as follows. ‘The magnitude of the shear force F found ftom equation (18) is checked against the maximum possible value (F).» dened as Foam =Fatan te... ay hier ¢, is the smaller of the interparticle friction angles of the two discs in contact and ¢ the Shaler oftheir cohesions.* Ifthe absolute value of (Fy as found from equation (18) s lager ‘than (Fw (Fy is Set equal to (Faq Preserving the sign obtained from equation (1), [ADISCRETE NU Once theno disc x, they ar force compon on disc x, Ei Motion ‘The velocit ‘obtained as f act on dise x « where Iu rep step 41, equa These equi may be used The new v by a further ‘The finite ai Body fore’ term Ma81 i the two com Damping Friction & contact is (F The prow to be referre relative velo the normal not applied "The coef by c, and 6, in the force andce summation is antheses are not ives the com- (as) as) At law to caleu- (16) an the sum of all as) ‘The sign con- 3: F,and F, he shear force Dane defined as as) tact and ¢ the (18) is larger tion (18). A DISCRETE NUMERICAL MODEL FOR GRANULAR ASSEMBLIES 33 ‘Once the normal and shear forces have been determined for each contact ofa dise, for example +x, they are resolved into components in the I and 2 directions. The sum of these contact, forces E Fy, and E Fiyn- The resultant moment acting in the counterclockwise direction and is found ai force components gives th "The resultant moments and forces acting on disc x are used with Newton's second law to determine the new accelerations Zand 6, Motion ‘The velocities £, and 0» used in the force-displacement law in equations (16) and (17) are obtained as follows. The current resultant force and the moment at time fy are assumed to act on disc x during the interval 4 from fy 0 fye4- Newton's second law applied to dise x is men’ E Foy oo es en here J.y represents the moment of inertia of disc x. Taking ¥, and 6, constant over the time step At, equations (20) and (21) lead to the following expressions for the velocities Gdyoy = Dna EZ Feomele dt fe @D Geadwes = Oca dn-s41Z Mallen dt + - @3) “These equations are applied to each disc in turn. These new values for the velocities now may be used in the force-displacement law and the cycle repeated for a new time increment. “The new values for velocities are used also to update the postions and rotations of the diss, by a further numerical integration dyer = ODr# (Deeg AE + Ge Oadnes = Cedvt Codnes dt» a (23a) “The finite difference equations (22), (228), (23), 23a) represent a time-centred system. Body forces, such as gravitational forces, may be incorporated, if so desired. In that case a term myo; is added to the force sum E Fi ia equation (22), where g, = (Bu 83) represents the two components of the acceleration vector due to the body force. Damping ‘Friction damping occurs during sliding when the absolute value of the shear force at any contact is (Fue ‘The program BALL further contains the option to include two forms of viscous damping to be referred to as contact damping and global damping, Contact damping operates on the relative velocities at the contacts and may be envisioned as resulting from dashpots acting in the normal and shear directions at the contacts. The viscous damping in the shear direction is not applied when sliding occurs. In this case friction damping alone operates. “The coefficients of viscous contact damping in the normal and shear directions are represented by c, and ¢,. If contact damping is taken into account, the damping forces must be included in the force-sums in equations (22) and (23), which become Gwe = Gdn -2+ E Feat Dewi mean At ears = Oerdn-s E Menlleady At ++ @5) a) and—S “ A. CUNDALL AND 0. , L,stRACK ‘where Z Duy; tepresents the sum of the components of the contact damping forces and where Muy now includes the contribution of the contact damping forces to the moment sum The slobal components Dj, are found from the normal (D,) and shear (D,) components of the ‘damping force atthe contacts, which are obtained from the following equations (sce equations (12) and (13)) Onn =e elt Sdn gs + (26) and dn = 6 = EL. F)y—s- Ooo Re + pKa] en ‘There isan error of half a time-step in the calculation of D, and D,, but its effects have been found to be negligible. In the current version of BALL, the contact damping coefficients ¢, and c, are taken to be Proportional tothe stffnesses k, and &, with proportionality constant p, oo On Global damping operates on the absolute velocities of the discs and is introduced in the cal= culations of motion. Global damping as incorporated in BALL may be envisioned as the elfect of dashpots connecting each particle to ground. ‘These dashpots operate both on the velocity vector components and on the rotational velocity. 1f global damping is included in addition to contact damping, the equations of motion 20) and (21), become (¢f. equations (24) and (25), n= By mk = 2 Font Din) Ch. 9) Thay = E Miyy~C%0 cy es G0) where C and C* are the coeficients of global damping operating, respectively, on 3, and 8. A central diference scheme is used to integrate equations (29) and (30), whereby velocities are evaluated halfway through the time step, ie. Gy = HEDn-4+ Gur Coy = HO dee Godel = BD Writing % a [C3)y+4~CDy-al/4t and Oy) a8 (Qua)ves—(e)n-sl/ At and using equation (1), equations (29) and (30) may be solved for (&)y.4 and (us)wes. Coeee= foe [1S S] 2 eat de fe S 4 . 2) Mey 2 ct a ct a (owes = Com Toi ‘fe Mody Al fi +i 3 - + G3) In BALL, the coeficients C and C* of global damping are taken to be proportional to mass ‘and moment of inertia, respectively and Cm amy Chm ake, oa avaton (4 equations (2) and (33) may besnfiied te ————_, Berm fede fied eerste Ay frre 65) 2 per 2 [DISCRETE NUMERIC These equation sranular material damping. The ut reach a state of € dladed, the assert ture from equilit It may be noted method. A non- "The numerical of the critical tin critical time ster fraction of hist on the basis of tt of stiffness k, fo ToiillustrateF specified by the dises are presen Input paramete The input p divided into t properties date ‘The geometr (the walls) and (L,2). A typie Xeon) (Bee Fig. passing throug endpoints A's tively. ‘The bounds rather than thNDO.D. L. smace forces and where noment sum. The ‘omponents ofthe ns (see equations Co) en effects have teen 4 ate taken tobe Cr) duced in the cal- ned as the eft hon the velosty + of motion, (0) 9) (0) 1 on &, and dy. Sy velocities are Gl) using equation ) es ‘tional to mass + 3 | DISCRETE NUMERICAL MODEL FOR GRANULAR ASSEMBLIES 35 Fig. 4. Possible wall definitions and Gcadnes {ede-f1-a$] 0 mann ZY frees} Ga “These equations for force-displacement and motion fully describe the BALL model of & ‘granular material. Energy is dissipated in the model only through friction, contact and global damping. The use of damping other than by friction is necessary in order thatthe assemblies, reach a state of equilibrium for all conditions. If neither contact nor global damping were in- cluded, the assemblies would not be able to reach exact equilibrium, but in practice the depar- ture from equilibrium may be made as small as desired by reducing the applied loading rate. It may be noted that the linear force-displacement law discussed above is not inherent in the method, A non-linear law, such as that of Herzian contact, could equally well be employed. “The numerical scheme will be stable only ifthe timestep 4r is taken as a fraction less than one of the critical time step, a result of the explicit nature of the model. An estimated value for the critical time step is determined in the program (see Cundall, 1978) and the user specifies the fraction ofthis time step thatis tobe used for the calculations, The critical time step is estimated on the basis of the single degree-of freedom system of a mass m connected to ground by a sprine Of stiffaessk, for which the critical time step equals 2(m/k). ‘Toillustrate how the program calculates and what effect variations inthe physical properties specified by the user have on the output ofthe program, the results of some simple tests on nine discs are presented after a brief discussion of the input parameters. Input parameters ‘The input parameters that must be specified to run numerical tests with BALL may be divided into two groups. One group consists of geometrical data and the other of physical properties data. ‘The geometrical data describe the positions and orientations of the straight rigid boundaries (the walls) and the positions and radii of the discs, with respect toa global co-ordinate system (1, 2). A typical wall is defined in terms of a fixed point P with co-ordinates w= Csr (wa) (6ee Fig. 4) and a unit vector e, = (cos a sin a) such that the wal i @ portion of the line passing through P and of the direction of e,. Scalar multiplies of e, A.) and ha define the wall endpoints A and B with co-ordinates xy. Xeovrthay & ANG. X= Xenrthay Gn FOSPEC™ tively. “The boundary conditions are ofthe strain-controlled type. The motion of the wallis specified rather than the forces applied to it. Movement ofthe wall is defined in terms of the velocity of