Incrementally non-linear plasticity applied to rock joint modelling

INTERNATIONAL JOURNAL FOR NUMERICAL AND ANALYTICAL METHODS IN GEOMECHANICSInt. J. Numer. Anal. Meth. Geomech. (2011)Published online in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/nag.1105

Incrementally non-linear plasticity applied to rock joint modelling

Jérôme Duriez1,*,†, Félix Darve1 and Frédéric-Victor Donzé1,2

1UJF-Grenoble 1, Grenoble-INP, CNRS UMR 5521, 3SRLab, Grenoble, France2Earth Science and Resource Engineering, CSIRO, Queensland Centre for Advanced Technologies,

Pullenvale, Australia

SUMMARY

Rock joint constitutive modelling is discussed through two new rock joint constitutive relations and adiscrete numerical model. Regarding the constitutive relations, we emphasise the number of ‘tensorialzones’, that is, domains of constitutive incremental linearity; they involve four zones for the first (called‘quadrilinear’) and an infinite number for the second one (called ‘incrementally nonlinear’). Using theseformulations, a large class of loading paths can be considered. Hardening through shearing and relationsbetween the normal and tangential directions of the joint (e.g., dilatancy) can be described. Their predictiveabilities are checked. Plastic features are included even if the relations are defined outside the elasto-plasticformalism. These relations obey, hence, the physical evidence as the plastic limit criterion and flow rule. Theflow rule is nonassociated, and the corresponding link with the nonsymmetry of the constitutive matrix isexamined. Comparisons between the two relations and the discrete numerical model, that is, a direct numer-ical simulation, which is fundamentally different, also are discussed within the context of infilled rock joints.Copyright © 2011 John Wiley & Sons, Ltd.

Received 8 June 2011; Accepted 6 September 2011

KEY WORDS: interpolation constitutive relation; rock joints; tensorial zone; discrete element method(DEM); associativity; incremental nonlinearity; flow rule

1. INTRODUCTION

Many rock slopes present a behaviour ruled by the rock joints’ mechanical properties. Hence, rockjoint constitutive modelling is the subject of this paper. Even when not planar, the geometry of theconsidered rock joints is supposed to be smooth enough so that a tangent plane can be taken intoaccount. Furthermore, only one tangent direction in the plane is considered, resulting in abidimensional mechanical state. This mechanical state is then defined by two scalar stresses—thenormal s (considered to be positive in compression) and the tangential t—and by two scalar relativedisplacements—the normal component u (considered to be positive in compression) and thetangential g. Those variables are linked by the constitutive relation of the joint. A direct formalism(s, t) = f (u, g) should be avoided, as the same ‘deformation’ state of the joint can lead to differentstress states. During a compression of rock joint, for example, because of the observed inelasticity[1], two different values of s can be obtained for the same value of u, whether this u-value isreached in loading or unloading. Consequently, we will use an incremental formalism

!ds ¼ ds; dtð Þ ¼

f~h

!dl

� �¼ f

~h du; dgð Þ, with the function f

~h depending on the state variables and the hardening parameter

~h.

From the rock joint’s rate-independent behaviour and Euler’s identity for homogeneous functions, we

obtain the existence of an elasto-plastic matrixM~h that depends only on direction

!d of

!dl

!d ¼ !

dl=!dl

�� (see Duriez et al. [2] for more details):

*Correspondence to: Jerome Duriez, 3SRLab, 38041 Grenoble Cedex, France.†E-mail: [email protected]

Copyright © 2011 John Wiley & Sons, Ltd.

J. DURIEZ, F. DARVE AND F.-V. DONZÉ

dtds

� �¼ M

~h

!d

� � dgdu

� �(1)

Different levels of model complexity are reached depending on how the matrixM~h evolves with

!d. If

M~h is constant for any given

!d, the relation is eventually incrementally linear and can describe only

elastic behaviours, which is not comprehensive. Tensorial zones [3,4] are then defined as sectionsof the (dg, du) plane in which M

~h is constant and equal to one unique matrix. The last example of

elasticity corresponds to a case with one tensorial zone and illustrates that a greater number oftensorial zones is required.

Usual rock joint constitutive relations are generally not consistent with this required incrementalformalism, with a high number of tensorial zones. Previous works by Bandis et al. [1] and Bartonet al. [5] have proposed relations directly linking stresses to relative displacements that do not obeythe incremental formalism mentioned previously. Incrementally linear constitutive relations for rockjoints also have been previously proposed [6,7]. Their limitation is that they present only onetensorial zone and, hence, are unable to reproduce the inelastic behaviours of rock joints, asexplained above. This inability can be corrected by elasto-plastic relations [8–10]. We also mentionthe work by Souley et al. [11], which improves on the results of Saeb and Amadei [7] from thispoint of view. Nevertheless, such relations still present only two tensorial zones, which is a limitednumber. Indeed, Hill’s results [12–14] suggest that a finite number of tensorial zones is adequate formetal crystals but not for geomaterials. The number of tensorial zones required corresponds to thedifferent types of plastic deformation mechanisms. In contrast to metals whose plasticity is ruled bythe sliding of dislocations, geomaterials seem not to have such special mechanisms [15,16].Simulating their behaviour thus requires the use of an infinite number of tensorial zones. Theincrementally non-linear, second-order (INL2) relations initially developed by Darve for soils [3,17]present this feature: the matrix M

~h depends on the direction

!d in a continuous manner. For a general

discussion of incremental nonlinearity, see Darve and Nicot [18]. Incremental nonlinearity has beenpreviously adapted to rock joint modelling [2,19]. The same construction will be used here, leadingto a new INL2 relation that improves on the previous work of Duriez et al. [2].

Section 2 then presents the construction of two rock joint constitutive relations: one INL2 and onequadrilinear (Quadri). The quadrilinear formalism corresponds to the octolinear relation (Octo)developed for soils in parallel with the INL2 relations. Even if the Octo formalism provides eighttensorial zones rather than an infinite number, its results are nevertheless comparable to those ofINL2 [20] (or section 5.1 of this paper). Here, the Quadri relation presents only four tensorial zones,which is still higher than the usual rock joint relations, but it allows the same easy analyticalanalysis, as Octo. Comparisons with the INL2 formalism is one of the goals of this paper, showingthat the results are generally close so that the analytical results obtained in Quadri case are usuallyvalid in the INL2 one.

These relations are calibrated in section 3. In that section, the relations are then validated by testingtheir predictive abilities. The calibration and validation of these relations rely on the numerical resultspresented by Duriez et al. [2]. These results were obtained from a rock joint numerical model usingdiscrete element methods (DEM), see Cundall and Strack [21]. In reference [2], the discrete resultswere compared with experimental ones. Thus, no other experimental result will be presented. Wenote that these discrete results have already been used in reference [2] to define an INL2 relation.The one presented here has some features in common with this previous relation, but significantimprovements have been made. We also note that the numerical model of reference [2] consideredinfilled rock joints. The framework of this previous work was indeed rock slope stability analysis,and in this case, such rock joints are critical because of their small shear strength [22–24]. Thisassumption of a special type of rock joint influences the values of the parameters, which will bedefined but not the way developments will be carried out. We mention here some general

considerations about constitutive relations!ds ¼ f

!dl

� �, which apply to any type of rock joint,

including the perhaps more general case of rough surfaces contact.The final sections discuss the proposed relations in the plasticity framework and examine their

theoretical basis. The plastic limit criterion is considered in section 4, the response envelopes in

Copyright © 2011 John Wiley & Sons, Ltd. Int. J. Numer. Anal. Meth. Geomech. (2011)DOI: 10.1002/nag

INL PLASTICITY APPLIED TO ROCK JOINT MODELLING

section 5 and the flow rule in section 6. Emphasis will again be placed on comparing the Quadri andINL2 formalisms and also with the discrete numerical results of the infilled rock joint model.

2. GENERAL DEFINITION OF TWO ROCK JOINT CONSTITUTIVE RELATIONS

Section 2.1 presents the construction of the two rock joint constitutive relations that are establishedoutside the usual elasto-plasticity framework. In this sense, no plasticity limit criterion, for example,is clearly stated. Nevertheless, to be realistic, the proposed constitutive relations have to describe thestress states that are limited by such plasticity limit criterion. This factor will be discussed in detailin section 4, but the fundamentals of the approach are presented in section 2.2.

2.1. Construction of the relations

The construction of a rock joint INL2 relation can be found in reference [2]. It leads to the followingequation that links the (incremental) stresses existing along the joint to the (incremental) relativedisplacements:

dtds

� �¼ 1

2Pþ þ P�ð Þ dg

du

� �þ 1

2ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffidu2 þ dg2

p Pþ � P�ð Þ dg2

du2

� �(2)

We recall, following [2], that the matrices P+ and P� are defined through eight moduli:

Pþ ¼ Gþg Gþ

uNþg Nþ

u

� �P� ¼ G�

g G�u

N�g N�

u

� �(3)

These moduli, Gþ=�g , Nþ=�

g , Gþ=�u and Nþ=�

u , characterise the behaviour of the joints along twocalibration paths:

• the first calibration path is a constant normal displacement (CND) shearing path that is controlledby (du= 0 ; dg = cst). From the resulting changes in s and t, four moduli are defined:

Gþg ¼ @t

@gu;dg>0G�

g ¼ @t@gu;dg<0

Nþg ¼ @s

@gu;dg>0N�g ¼ @s

@gu;dg<0

(4)

In these definitions, dg> 0 corresponds to the shear loading, whereas the unloading is associated

with dg< 0. The Gþ=�g moduli are similar to classical tangential rigidities, and the Nþ=�

g modulidescribe the dilatants feature of the joint. Figure 1 illustrates the definitions of Equation (4).

• a second calibration path is the constant tangential displacement (CTD) path, which is controlledby (du= cst ; dg = 0). In this case, four other moduli can be defined:

Figure 1. Definition of four moduli from an arbitrary constant normal displacement (CND) path behaviour.



Gþu ¼ @t

@ug;du>0G�

u ¼ @t@ug;du<0

Nþu ¼ @s

@ug;du>0N�u ¼ @s

@ug;du<0

(5)

The Gþ=�u moduli describe how the tangential stress t evolves during changes in the normal

displacement u. Indeed, Duriez et al. [2] discusses how compressions can affect the values of t incases where the joint has been previously sheared. The Nþ=�

u moduli correspond to the classicalnormal rigidities. Figure 2 illustrates Equation (5).

We recall that Equation (2) corresponds to Equation (1), with continuous changes inM~h with respect

to!d. It also corresponds to an interpolation. In this case, the behaviour of the joint along any loading

path is determined by the behaviours corresponding to the calibration paths through a nonlinear(quadratic) interpolation (see, e.g., reference [25]). The other constitutive relation, Quadri, is definedif a piecewise linear interpolation is used. It comes with the matrices P+ and P� in this case:

dtds

� �¼ 1

2Pþ þ P�ð Þ dg

du

� �þ 12

Pþ � P�ð Þ dgj jduj j

� �(6)

To illustrate Equations (2) and (6), we emphasise that, for the (du= 0 ; dg 6¼ 0) loading, we obtain:

f dt ¼ Gþg þ G�

g

� � dg2þ Gþ

g � G�g

� � dg

2ffiffiffiffiffiffiffidg2

p for INL2 case; Equation 2ð Þ

or dt ¼ Gþg þ G�

g

� � dg2þ Gþ

g � G�g

� � dgj j2

for Quadri case;Equation 6ð Þ, dt ¼ Gþ

g dg if dg > 0 and dt ¼ G�g dg if dg < 0; for both cases

(7)

Equation (7) obviously corresponds to the definitions of the moduli, but it also expresses a classicalelasto-plastic behaviour, with different loading and unloading stiffness.

The calibrations of the two constitutive relations, Equations (2) and (6), are identical; thus, theresponses of these two relations for calibration paths are identical. For other loading paths, bothpredictions are close, with the INL2 formalism being generally more precise (comparisons will bepresented in this paper). However, the Quadri case has the advantage of making analyticalcomputations possible. If constant signs for dg and du are assumed, which means to consider agiven tensorial zone, then the Quadri relation is indeed linear. Expressions for the correspondingmatrices M

~h are presented in Table I. As we will see, the analytical results obtained in this way can

be extended to the INL2 case.

2.2. Link with a plasticity limit criterion

As stated in the introduction, the use of Equation (2) or (6) do not require the use of a plasticity limitcriterion. However, such criteria are physical evidence and have to be fulfilled by any constitutive

Figure 2. Definition of four other moduli from an arbitrary constant tangential displacement (CTD) pathbehaviour.


Table I. Constitutive matrices for the different tensorial zones of Quadri relation.

Mh dg< 0 dg> 0

du> 0 G�g Gþ

uN�g Nþ

u

� �¼ M�þ Gþ

g Gþu

Nþg Nþ

u

� �¼ Mþþ

du< 0G�

g G�u

N�g Nu�

� �¼ M�� Gþ

g G�u

Nþg N�

u

� �¼ Mþ�


relation. For rock joints, it is always observed, for example, that the tangential stress t reaches a peakand/or a plateau at the end of a constant normal load (CNL, with s= cst) shearing [1,7,24,26,27]. Theconsequences for the definition of constitutive relations are presented here.

Let us consider any incrementally linear rock joint constitutive relation, as defined by followingEquation (8):

dtds

� �¼ Gg Gu

Ng Nu

� �dgdu

� �(8)

This relation holds for our Quadri relation in a given tensorial zone. In this framework, peaks orplateaus in t for the CNL tests can be expressed mathematically. As the CNL conditions alreadyimply that ds= 0, obtaining dt= 0 at t peak or t plateau implies that

!d s ¼ M

!dl ¼ !

0;with!dl 6¼ !

0 , det Mð Þ ¼ 0 , GgNu � GuNg ¼ 0 , GgNu ¼ GuNg (9)

Let us note that det(M) = 0 corresponds exactly to the general definition of a plastic limit criterion inelasto-plastic theory.

The previous work by Duriez et al. [2] was incomplete in this aspect. A Mohr–Coulomb plasticitylimit criterion was considered to limit the stress states: t⩽ stan(’). However, the moduli expressionswere not consistent with this plasticity limit criterion regarding Equation (9), as Figure 3 illustrates.The evolution of the determinants of the two matrices corresponding to dg> 0, taken as examplesamong four, are plotted according to t from t = 0 to t = tmax = stan(’). For each determinant of thetwo matrices, different constant values of s (1; 2 and 5MPa) are considered. Thus, it illustrates that,with moduli such as those defined in reference [2], the determinant of the matrices can cancel both

0 0.5 1 1.5 2 2.5 3−100

0

100

200

300

400

τ (MPa)

det(

M)

(MP

a/m

m)2

dγ>0;du<0dγ>0;du>0

σ = 1 MPa

σ = 2 MPa

σ = 5 MPa

Figure 3. Determinant of two constitutive matrices of reference [2], according to t (from nonsheared tocompletely sheared states), under different values of s.



on the plasticity limit criterion (maximum value of t in the figure) and before the plasticity limitcriterion, as in the case of the matrix corresponding to dg> 0, du< 0 (the typical case for a CNL testduring the dilatant phase). This result will be improved in the present work.

To conclude this section, we note that Equation (9) can be physically interpreted. We canconsider Nu and Gg as diagonal rigidities in the sense that they link together the normalvariables (ds with du) or the tangential variables (dt with dg) and, hence, Gu and Ng asnondiagonal ones. Equation (9) thus states that on the plasticity limit criterion, the (product of)nondiagonal rigidities are as great as the (product of) diagonal ones. More generally, a linkbetween these diagonal and nondiagonal terms is suggested through this equation. This resultwill be used in following calibration of our relations (Quadri and INL2). Such links also havethe advantage of reducing the number of parameters.

3. CALIBRATION AND VALIDATION OF THE RELATIONS

Calibration and validation of the INL2 relation, not of the Quadri relation, have been proposed inreference [2]. As explained in section 1, past work has focused on granular infilled rock joints.Thus, a numerical model using DEM [21] (with the Yade code [28]) has been used to obtain asufficient database for our calibration paths (CND and CTD). The choice of DEM was motivated bythe presence of the granular filler, as many works show the quantitative and qualitative abilities ofDEM to simulate granular media [29–33]. In reference [2], the DEM results were compared withsome new experimental results; such comparison will therefore not be presented again.

In the present paper, the same DEM results will be used. However, for the reasons discussedpreviously in section 2.2, a new calibration will be performed in section 3.1. INL2 or Quadrirelations present no differences in the calibration step. For validation, section 3.2 will examine thepredictive abilities of these two relations and illustrate their differences.

3.1. Calibration

The Figure 4 shows the behaviour of the joint along CND paths, as determined by the DEMsimulations [2] taken as reference. Different CND paths were performed, with different initial valuesfor s0, in loading and unloading. Equation (10) proposes new definitions of the corresponding

moduli Gþ=�g ;Nþ=�

g

� �so that the constitutive relations provide the same response.

0 2 4 6 8 10 120

10

20

30

40

50

γ (mm) γ (mm)

σ (M

Pa)

σ0=1 MPa

σ0=5 MPa

σ0=10 MPa

σ0=20 MPa

(a) ( )

0 2 4 6 8 10 120

5

10

15

20

τ (M

Pa)

(b) ( )

σ0=1 MPa

σ0=5 MPa

σ0=10 MPa

σ0=20 MPa

Figure 4. Reference CND paths behaviour, from reference [2], obtained using discrete element method(DEM) computations.



Nþg ¼ f N0

g < 0 if t=s < tan ’cð ÞNþg dil > 0 otherwise

G0g ¼ G0

g � G0g � Gf

g

� � 1� e�

t=stan ’ð Þ

1� e�1where Gf

g ¼ Nþg dil tan ’ð Þ

N�g ¼ f N0

g < 0 if t=s < tan ’cð ÞG0

g

tan ’ð Þ > 0 otherwise

G�g ¼ G0

g

(10)

The Nþg moduli takes first negative and then positive values, to describe the observed contractant–

dilatant transition, as shown in Figure 4(a). TheGþg moduli evolves continuously during shearing from

an initial value of G0g to a final value of Gf

g, as suggested by Figure 4(b). As discussed in reference [2],the shearing is taken into account through the ratio of stresses t/s (e.g., rather than the displacement g).The final value, Gf

g , is expressed according to a parameter related to Nþg , Nþ

g dil (because of thediscussion of the plastic limit criterion of section 2.2) and another parameter, the angle ’. The N�

g

unloading moduli is expressed equal to the Nþg value in the contractant domain; otherwise, a link

with Gg (and ’) is still used. Finally, the G�g unloading moduli is considered to be equal to the

initial value of the corresponding loading moduli Gþg .

Figure 5 compares the DEM results with the INL2 or Quadri results, with such expressions ofmoduli. The stress levels observed in the DEM results are not reproduced with the constitutiverelations. Such levels are considered to be artefacts caused by a numerical sample being no longerhomogeneous for high values of g. This issue will be discussed more carefully in section 6. With theexception of these levels, a sufficient agreement is reached, which supports Equation (10).

Second, Figure 6 presents the reference behaviour for the CTD paths. The numerical sample wassubjected to a first CTD compression until a value of s � 10 MPa was reached. It was then shearedunder constant s until different values of g were obtained (the effects of these two first steps donot appear in Figure 6). From these different initial states, different CTD paths were simulatedfor loading (du> 0) and unloading (du< 0). The corresponding variations in s and t are plottedon Figure 6.

Equation (11) proposes new definitions of the moduli corresponding to the curves in Figure 6:Gþ=�

u , and Nþ=�u .

0 2 4 6 8 10 120

10

20

30

40

50

γ (mm)0 2 4 6 8 10 12

γ (mm)

σ (M

Pa)

DEMINL2

0

5

10

15

20

τ (M

Pa)

σ0=1 MPa

σ0=5 MPa

σ0=10 MPa

σ0=20 MPa

σ0=1 MPa

σ0=5 MPa

σ0=10 MPa

σ0=20 MPa

DEMINL2

(a) ( ) (b) ( )

Figure 5. Calibration of INL2 or Quadri relations for CND paths (here, INL2 =Quadri).


−0.6 −0.4 −0.2 0 0.2 0.4 0.60

5

10

15

20

u (mm)

σ (M

Pa)

sheared

Initial statesincreasingly

γ0=0.5mmγ0=2mmγ0=4.5mmγ0=6.5mm

γ0=1.5mmγ0=3mm

(a) ( )

−0.4 −0.2 0 0.2 0.4 0.6

0

2

4

6

u (mm)

τ (M

Pa)

Initial statesincreasinglysheared

(b) ( )

γ0=0.5mmγ0=2mmγ0=4.5mmγ0=6.5mm

γ0=1.5mmγ0=3mm

Figure 6. Reference CTD paths behaviour, from reference [2].


Nþu ¼ N

ts

� � ss0

� �13

with Nts

� �¼ 20N0

u � Nfu

19� N0

u � Nfu

1920

t=stan ’ð Þ

N�u ¼ Nd � Nþ

u

Gþu ¼ FP

u

tan ’ð Þts

ss0

� �13

G�u ¼ GM

u

tan ’ð Þts

ss0

� �13with GM

u ¼ Nd � Nfu � tan ’ð Þ

(11)

In the case of normal stiffness Nþ=�u , the reference behaviour shows, consistent with experimental

results [1,7], an increase in the normal stiffness according to s and a decrease of this stiffness whileshearing. Thus, the corresponding Nþ

u moduli is defined as increasing with respect to s and, throughfunction N(t/s), decreasing according to t/s. The use of the function N leads to values of Nþ

u , fors= s0, equal to N0

u in a virgin state (with no shearing), and equal to Nfu if complete shearing is

achieved (t/s= tan(’)). Section 4 will indeed show that tan(’) is the maximum value of the ratio t/s.The corresponding unloading moduli N�

u is taken to be proportional to Nþu , through parameter Nd.

The Gþ=�u moduli are then defined as increasing with respect to both shearing (the t/s ratio) and

normal stress s. A parameter GPu appears in the Gþ

u definition. It corresponds to the value of Gþu

under the completely sheared state: t = tmax = stan(’), for s=s0. For the definition of G�u , a link is

set between G�u and N�

u , following section 2.2.Figure 7 shows both the DEM results and the INL2/Quadri relations (Equations (2) and (6)) for such

CTD paths. Good agreement is reached, which completes the calibration of our relations. Theparameters and their retained values are summarised in Table II. We finally obtain nine independentparameters (the value of s0 is arbitrary and linked to the values of N0

u , Nfu , for example).

3.2. Validation

To validate one or two of the constitutive relations, loading paths other than CND or CTD ones areconsidered. The DEM results of Duriez et al. [2] are still considered to be the reference. In thisprevious paper, the CNL and constant normal stiffness (CNS) paths were addressed. The


−0.4 −0.2 0 0.2 0.40

5

10

15

20

u (mm)

σ (M

Pa)

Increasinginitialshearing

−0.4 −0.2 0 0.2 0.4 0.60

2

4

6

u (mm)

τ (M

Pa)

Increasingshearing

DEMINL2

DEMINL2

(a) ( ) (b) ( )

Figure 7. Calibration of INL2/Quadri relation for CTD paths (here, INL2 =Quadri).


corresponding results will be compared with the predictions of Equations (2) and (6), with moduli suchas those used here, in the Equations (10) and (11).

The results of the discrete model of reference [2] and its predictions for the constitutive relations arecompared in Figure 8 for CNL paths under different values of s: 1, 5, 10 and 20MPa.

Concerning the t(g) curve, Figure 8(a) shows that the INL2 and Quadri predictions are close to eachother and to reference discrete results. In the case of u(g) in Figure 8(b), the differences between thetwo predictions appear. Compared with DEM, the INL predictions are very efficient for middlevalues of s (5 or 10MPa) and a little less for extreme values of s (1 or 20MPa). The Quadripredictions are further from those of DEM, with the exception of the s=1MPa case. In general, theQuadri predictions correspond to a less dilatant behaviour than do the INL ones.

Constant normal stiffness paths are then considered. Such loadings are shearings for which attentionis paid both to changes in s and u. The test is then monitored so that ds/du= w during the entire test,where w is a chosen constant. Figure 9 shows the comparison between the discrete results and the INL2or Quadri predictions for three different CNS tests. The three tests start from the same value of s(5MPa) but correspond to different values of w: 10, 40 and 120MPa/mm. Again, the INL2behaviour here is more dilatant than that of the Quadri (see the s(g) and u(g) curves in Figures 9(c)and 9(b)), and the INL2 predictions are still generally more efficient than the Quadri ones, even ifthe differences are not as important. As differences would be especially so small in Figure 9(d), theQuadri results are not plotted in this subfigure so that the curves will remain readable. Finally, theseresults lead us to consider our constitutive relations to be reasonably validated.

Now that calibration and validation have been performed, a plasticity analysis of the two relationswill be considered, focusing first on plasticity limit criterion.

4. PLASTICITY LIMIT CRITERION OF THESE CONSTITUTIVE RELATIONS

Following the considerations of section 2.2, the existence of a plasticity limit criterion is discussed. Aparameter ’ was introduced in the expressions for some of the moduli. This section will show that thisparameter corresponds to a friction angle in a Mohr–Coulomb criterion that limits the stress states

Table II. INL2/Quadri relations parameters.

(GPa/m) (MPa) (�)

N0g Nþ

g dil G0g N0

u Nfu GP

u Nd s0 ’ ’c

�2.4 3.0 3.6 20 8.0 2.4 2.0 1.0 29 12


0 2 4 6 80

2

4

6

8

10

12

γ (mm)

τ (M

Pa)

σ0=1 MPaσ0=5 MPaσ0=10 MPaσ0=20 MPa DEM

INL2Quadri

(a) ( )

0 2 4 6 8−1

−0.8

−0.6

−0.4

−0.2

0

γ (mm)

u −

u0

(mm

)

(b) ( )

σ0=1 MPaσ0=5 MPaσ0=10 MPaσ0=20 MPa

DEMINL2Quadri

Figure 8. Predictions of the constitutive relations towards discrete results (from reference [2]) for constantnormal load (CNL) loading paths.

0 2 4 6 8 10 120

2

4

6

8

10

12

14

γ (mm)

τ (M

Pa)

DEMINL2Quadri

χ=10 MPa/mmχ=40 MPa/mmχ=120 MPa/mm

(a) ( )

0 2 4 6 8 10 12

−0.6

−0.4

−0.2

0

0.2

γ (mm)

u (m

m)

(b) ( )

0 2 4 6 8 10 12

5

10

15

20

25

30

γ (mm)

σ (M

Pa)

(c) ( )

5 10 15 20 25 300

5

10

15

σ (MPa)

τ (M

Pa)

DEMINL2Quadri


DEMINL2

DEMINL2

Quadri


(d) ( )


igure 9. Predictions of the relations towards discrete results (from reference [2]) for constant normal stiff-ness (CNS) loading paths.


F

described by our relations. The Quadri and INL2 cases will be successively debated. The existence ofthe plasticity limit criterion for the discrete model was discussed previously [2].



4.1. Quadri case

Let us assume a stress state located on this supposed Mohr–Coulomb criterion: (s, t =stan(’)). Fromthis state, any loading

!dl is considered, and it is defined by its direction θdl (see Figure 10(a)). The

direction of the corresponding response!ds is denoted by θds (see Figure 10(b)). Because of

Equation (1), the direction θds depends only on the direction θdl. The Mohr–Coulomb criterionwould then be violated if there exists some θdl leading to θds2 [’ ;’ + 180�] = [29� ; 209�], in ourcase where ’= 29� (see Table II).

Figure 11 examines this question. Thanks to the use of Quadri relation, Equation (6), θds can becomputed according to θdl. For the stress states of the Mohr–Coulomb criterion, the correspondingrelation still depends on s, and the two values (1 and 10MPa) are thus considered in Figure 3.Figure 11 shows that values of θds between ’= 29�and 180 +’= 209�are never reached. Theplasticity limit criterion defined by tmax = stan(’) seems here to be never violated. Moreover,Figure 3 illustrates that there are some loading directions that lead to the responses following theMohr–Coulomb criterion; these loadings all imply the compression unloadings du< 0(θdl2 [180� ; 360�]). The discrete results of Duriez et al. [2] (see also Figure 23) have indeed shownthat the CTD unloadings (θdl= 270�) from completely sheared states lead to decreases in thestresses, with the stress states remaining consistent with the Mohr–Coulomb criterion, whereas anyloading with the compression du> 0 (θdl2 [0� ; 180�]) carries the stress state strictly inside theplasticity limit criterion.

Figure 10. Definition of angles θdl (in displacement plane) and θds (in stress plane).

0 90 180 270 3600

29

90

180

209

270

360

θdl (deg)

θdσ

(deg

)

σ=1 MPa;τ=τmax

σ=10 MPa;τ=τmax

A

B

Figure 11. Quadri case: link between displacement loading and stress response directions. For two stressstates on plasticity limit criterion, AB discontinuity is linked to the existence of a plastic limit criterion of

Mohr–Coulomb type.



The properties of the function θds= f(θdl) are linked with the cancellation of the determinants of theconstitutive matrices, as described in section 2.2. Figure 12 shows the evolution of the determinants ofthe four matrices from the Quadri case. The determinants are generally decreasing (from their initialpositive values) during the shearing. A discontinuous change is observed that corresponds to the

discontinuous change of the Nþ=�g moduli at the contractant–dilatant transition. Among the four

matrices, the two corresponding to the tensorial zones with du< 0 vanish exactly on the plasticitylimit criterion. Compared with Figure 3, which considered other expressions for the moduli, thecancellations no longer occur before the plasticity limit criterion. In fact, as the matrices are definedinside a tensorial zone, it must be verified that the eigenvectors of the eigenvalue 0 (obtainedbecause of the determinant cancelling) belong to the correct tensorial zone—we note already thatsuch eigenvectors correspond to flow rule, as it will be discussed in section 6. This verification isperformed in Figure 13. In this way, we verify that the eigenvectors of eigenvalue 0 do not belongto the right tensorial zone for (dg< 0, du< 0), as they correspond to (dg> 0, du< 0) or(dg< 0, du> 0). Hence, the corresponding matrix for this tensorial zone is finally not to beconsidered with a vanishing determinant. By contrast, cancellation of the determinant of thecorresponding matrix is confirmed for (dg> 0, du< 0). The eigenvectors of the matrix belong to thetensorial zone in which the matrix exists.

0 0.5 1 1.5 2 2.5 3 3.50

50

100

150

200

250

τ (MPa)

det(

M)

(MP

a/m

m)2

σ = 1 MPa

σ = 2 MPa

σ = 5 MPa

dγ>0;du<0

dγ>0;du>0

dγ<0;du<0

dγ<0;du>0

Figure 12. Quadri case: determinants of constitutive matrices according to t for different values of s.

−1 −0.5 0 0.5 1−1

−0.5

0

0.5

1

dγ

du

1 MPa2 MPa5 MPa

(a) ( < 0, < 0) tensorial zone

−1 −0.5 0 0.5 1−1

−0.5

0

0.5

1

dγ

du

1 MPa2 MPa5 MPa

(b) ( > 0, < 0) tensorial zone

Figure 13. Eigenvectors of eigenvalue 0, for both matrices corresponding to du< 0 (with dg> or< 0). Forthis vanishing eigenvalue, directions have to be considered, not only vectors.



Because of the expressions for moduli in Equations (10) and (11), such cancellations are obtainedfor any value of s. As we obtain in our case, on the plasticity limit criterion:

det Mþ�ð Þ ¼ Gþg N

�u � Nþ

g G�u ¼ Gf

gNdNfu

ss0

� �13 � Nþ

g dilGMu

ss0

� �13

¼ ðNþg diltan ’ð ÞNdNf

u � Nþg dilNdNf

utan ’ð ÞÞ ss0

� �13 ¼ 0 8s

(12)

These results prove the existence of a plasticity limit criterion (of the Mohr–Coulomb type withoutcohesion), outside of which, stress states cannot be obtained with our Quadri relation. This criterion isnot directly stated but appears through the retained expressions for the moduli. The definitions allowthe required cancellations of the determinants of the constitutive matrices in the criterion. Moreover,our results illustrate the fact that limit stress states can be obtained only with a shear loading(dg≥ 0) that is linked with dilatancy (du< 0). Whereas other loadings (e.g., loading compressionsdu> 0) move the rock joint stress state further inside the plasticity limit criterion. This result isconsistent with a nonvanishing determinant of the constitutive matrix (see also next section 6.5).

4.2. INL2 case

For the INL2 case, such an analysis using piecewise constant constitutive matrices cannot beperformed. However, stress states also seem to be limited in this case: see for example Figure 8(a),in which the INL2 t(g) curves for the CNL tests also level off (without any correction). Thethreshold obtained is close to the one for the Quadri curves. In addition to the CNL tests consideredin Figure 8, nine other CNL tests were carried out with INL2 relation using other s values in therange [1MPa;20MPa]. For all of these CNL tests, we observe that the maximum value of t, tmax,corresponds to values of arctan(tmax/s) remaining in the range [29.4�;29.5�].

To determine if this condition remains valid for any loadings other than the CNL ones, we performthe same analysis that led to Figure 3. The final stress states (i.e., t thresholds) for the CNL loadingsunder 1 and 10MPa (performed using the INL2 relations) are considered. From these mechanicalstates, a displacement probe test is performed in the INL2 framework. The stress response directionθds can then be plotted according to θdl in Figure 14. The major point of Figure 14 is that therealso is no stress response increment for the INL2 case in the [29;209] range, which would bring thestress state outside the plasticity limit criterion. We note that there might be an approximately 0.5�

0 90 180 270 3600

29

90

180

209

270

360

θ dl (deg)

θ dσ

(de

g)

σ=1 MPa;τ=τmax

σ=10 MPa;τ=τmax

A

B

Figure 14. INL2 case: link between displacement loading and stress response directions. For two stressstates, AB discontinuity is linked to the existence of a plastic limit criterion of Mohr–Coulomb type.



error on these values, as the value of the observed friction angle in the INL2 case is approximately29.5� rather than exactly 29�.

Finally, the existence of a plasticity limit criterion also is obtained for the INL2 relation. It can beconsidered to be identical to the one for the Quadri relation. Both are obtained from the moduliexpression. Using the response the response envelopes, following section 5 illustrates the changes inthe rock joint behaviour when the plastic limit criterion is approached.

5. RESPONSE ENVELOPES

Response envelopes studies [34] allow checking the consistency of proposed constitutive equations[20]. Indeed, in the case of incrementally piecewise linear constitutive relations, the elasto-plasticmatrix is changing suddenly from one tensorial zone to another (we recall that, by definition, theconstitutive relation inside a tensorial zone is linear and thus characterised by a given matrix).However, from physical and experimental points of view, the material response must be continuouswhen the loading direction is passing through the border between two tensorial zones. Thiscondition is the so-called ‘continuity condition’ (see reference [4] for details). For elasto-plasticconstitutive relations, it is possible to demonstrate that the continuity condition is always fulfilled ifthe consistency condition of the elasto-plastic theory is properly written. However, in this paper, theconstitutive relations have not been developed in a classical elasto-plastic framework, and thus, thecontinuity condition must be specifically verified.

From an analytical point of view, we consider the Quadri case, Equation (6), and Table I. As anexample, we focus on the tensorial zone (dg< 0, du> 0), with the elasto-plastic matrix M equal toM�+, and the tensorial zone (dg> 0, du> 0) with the M++ matrix. The frontier between these twozones is the half straight line in the (dg, du) plane whose equation is (dg= 0, du> 0). The firstcolumn of the M matrix is changing through this frontier, from M� + to M++. In a consistent manner,this first column is multiplied by dg= 0 to determine the material response, which is thus perfectlycontinuous.

This analytical reasoning can be graphically confirmed by considering the response envelopes that

illustrate the material response vectors (!ds or

!dl ) for a given state by considering the loading vectors

(!dl or

!ds) in all the directions θdl or θds (see previous Figure 10). Hence, the response envelopes

are presented for our relations. Comparisons are performed between the Quadri and INL2 cases.Then, the influences of the normal stress value and shearing state are investigated. The shearing stateis described by the variable fmob= arctan(t/s). No shearing corresponds to fmob=0, and completeshearing corresponds to fmob=’= 29� (see section 4). The displacement and stress probes areimposed; in both cases, 3600 incremental displacement loadings are considered, resulting in θdl orθds covering (0�;360�) by steps of 0.1�. The norms of the stress or displacement probes

!ds

�� ¼100kPa; or!dl

�� ¼ 10�5m� �

are arbitrary and do not affect the shape of the envelopes here.

5.1. Comparison between Quadri and INL2 cases

A comparison between the envelope responses of our two relations is shown in Figure 15. For thisFigure, the displacement probes, from two different mechanical states, are considered. A constantvalue for s (10MPa) is used, but two different values of t, that is, two different values of fmob, areconsidered. This Figure 15 shows that the Quadri and INL2 relations have similar responseenvelopes. Slight differences appear only for the sheared states of the joint. For this reason, only theINL2 relation is considered in sections 5.2 and 5.3.

We remark that an elastic behaviour is characterised by an ellipse whose centre corresponds to theplane origin (ds= 0, dt= 0). For an incrementally piecewise linear elasto-plastic relation, the responseenvelope is constituted by arcs of ellipses that form a continuous diagram if the constitutive relation iscontinuous. The elasto-plastic natures of the proposed constitutive relations and their continuities areclearly visible in Figure 15.

As a comparison with previous similar relations, Figure 16 uses data from reference [20] to comparethe response envelopes of two INL2 and Octo relations that were calibrated on soils. We recall that the


−800 −600 −400 −200 0 200 400

−400

−200

0

200

400

dσ (kPa)dτ

(kP

a)

φmob

= 5o

φmob

= 27o

INL2

Quadri

Figure 15. Comparison of envelope responses for INL2 and Quadri relations, for s=10MPa and two differentvalues of t.


Octo relation corresponds to the Quadri relation but in three dimensions (3D; it then presents eighttensorial zones). In reference [20], stress probes were performed under axisymmetric conditions,from different stress states (s1 ; s2 = s3 ; s3). In this case also, the differences between Octo (which isthus compared with the Quadri case) and INL2 relations appear significant only if the sample issufficiently sheared. The differences in this case are greater than those in the case of the rock jointrelations.

5.2. Influence of s

Figure 17 considers four response envelopes of the INL2 relation for four different mechanical states,with two different values of s, under two shearing states (two different fmob). Whatever the shearingstate, increasing values of s change only the size of the response envelopes and not their shape.

5.3. Influence of shearing

The influence of previous shearing on the response envelopes is then studied alone, with a constantvalue for s (10MPa). Figure 18 considers displacement and stress probes for several fmob values.Contrary to the displacement case, no stress probe can be performed under fmob= 29�. Indeed,

−0.02 0 0.02−0.04

−0.02

0

0.02

0.04

20.5 ε2 = 20.5 ε3 (%) 20.5 ε2 = 20.5 ε3 (%)

Octo

INL2

(a) 1 3 100 kPa

−0.4 0

0

0.2

0.4

0.6

ε 1 (%

)

ε 1 (

%)

Octo

INL2

(b) 1 400 kPa; 3 100 kPa (near failure)

Figure 16. Comparison of response envelopes for simulations with Octo and INL2 relations, for soils’ case,after data of reference [20].


−800 −600 −400 −200 0 200 400−600

−400

−200

0

200

400

dσ (kPa)dτ

(kP

a)

σ = 3 MPaσ = 18 MPa

φmob

= 15o

φmob

= 28o

Figure 17. INL2 response envelopes: influence of s.


section 4.2 showed that, for these mechanical states, the set of stress directions θds directed outside theMohr–Coulomb limit line is never reached, for any value of θdl. These directions cannot then beconsidered, and complete stress probe cannot be executed. That being said, the evolutions of thestress response envelopes during shearing appear clearly in Figure 18(a). From an initial curve thatis close to an ellipse, the stress envelopes change until presenting, under the Mohr–Coulombcriterion, a straight section that corresponds to the plasticity limit criterion (θds= 209� for thisstraight section). The same trends can be observed for the displacement response envelopes inFigure 18(b), with the envelopes particularly growing in the flow rule direction as the shearingincreases.

6. FLOW RULES OF THE RELATIONS AND THE DISCRETE MODEL

When a limit stress state is reached on the plastic limit criterion ð !ds

�� ¼

!0, e.g., CNL shearing at the

end of the shearing), the relative displacements still evolve. However, their direction is fixed by theflow rule of the rock joint. A precise definition of this flow rule will be proposed before it is appliedto our relations and our discrete model. Comparisons and concluding remarks about nonassociativityand singularity of the flow rule are then presented.

−800 −600 −400 −200 0 200 400

−400

−200

0

200

400

dσ (kPa)

dτ (

kPa)

φmob

=5o

φmob

=20o

φmob

=25o

φmob

=29o

(a) Displacement probes

−4 −2 0 2 4 6 8

−5

0

5

x 10−5

x 10−5

dγ (m)

du (

m)

φmob

=5o

φmob

=20o

φmob

=25o

(b) Stress probes

Figure 18. INL2 response envelopes: influence of shearing (for s= 10MPa).



6.1. Definition of the flow rule

Classical elasto-plastic relations define elastic and plastic parts of strains (relative displacements inrock joints case). In this framework, plastic strains, once they appeared, have a direction imposed bythe material flow rule. The Quadri andINL2 relations do not distinguish plastic or elastic strains(relative displacements). Nevertheless, we can define a flow rule from the displacement increments

that occur when the stresses reach the plastic limit criterion: the vector!dlp ¼ dup; dgpð Þ . As

discussed by Darve and Nicot [35], this definition is close to the classical one; as on the plastic limitcriterion, the elastic part of the deformation becomes negligible. In our case, the flow rule controlsthe ratio dup/dgp through a dilatancy angle c, such that

tan cð Þ ¼ � dup

dgp(13)

Figure 19 illustrates this definition. The (s, t) and (u, g) planes are superposed on the figure, as it isrequired when flow rules are tackled. The angle c can be measured from the t (or g)-axis to the vector!dlp. For c=’, we verify that we would obtain an associated behaviour:

!dlp would be orthogonal to

plastic limit criterion.The dilatancy angle c can, for example, be observed in Figure 8(b) as the angle between the u(g)

curves and the g�axis, for the section of the curves corresponding to the t plateau.

6.2. Flow rule for Quadri case

Linear expressions in the Quadri case allow the flow rule to be easily determined. We solve equation

ds ¼ M!dl ¼ !

0, with!dl unknown and M such that det(M) = 0 (the plastic limit criterion) and determine

the direction of the set of solutions that is obtained in this case. This is performed in following Equation(14), in which moduli of the (dg> 0, du< 0) tensorial zone are considered. Section 4.1 showed that the

vectors!dl from the other quadrants cannot be solutions to this equation, implying that the flow rule

belongs to the (dg> 0, du< 0) tensorial zone.

f dt ¼ Gþg dgþ G�

u du ¼ 0ds ¼ Nþ

g dgþ N�u du ¼ 0 , du

dg¼ �Gþ

g

G�u

¼ �Nþg

N�u

, tan cð Þ ¼ Gþg

G�u

¼ Nþg

N�u

(14)

Obviously, equality of the last two terms in Equation (14) corresponds to cancellation of thedeterminant. With the expressions for the moduli proposed in Equations (10) and (11), we finallyobtain

Figure 19. Rock joint flow rule: illustration.



tan cð Þ ¼ Nþg dil

NdNfu s=s0ð Þ1=3

(15)

Such a flow rule is plotted according to s in Figure 22 (appearing with other results in section 6.4).Equation (15) indicates already that the dilatancy angle c decreases according to s and to a parameterlinked with normal rigidity Nf

u

� , whereas increasing the dilatant feature of the joint (Nþ

g dil parameter)allows c to increase.

6.3. Flow rule for INL2 case

For the INL2 case, the previous analytical development is no longer possible. However, the values ofc(s) can be measured in two ways.

First, the slopes of u(g) curves for the CNL tests, once on the t plateaus, are measured from resultspresented in Figure 8(b). As explained before, such slopes depend directly on c through Equation (13).

Another method for determining c value leads to the same results. It relies on ‘proportionaldisplacement loading paths’ that are performed using INL2 relation. Such paths are governed bydu/dg =R, du�Rdg = 0.

To understand the behaviour for such paths on the plasticity limit criterion, it is only necessary tocompare the volume variation rate imposed by the loading path (through the constant R) to thematerial volume variation rate (which is issued from the dilatancy angle, c). If the loading pathvolume variation rate is higher (i.e., more dilatant) than the material volume variation rate, stressesdecrease along or close to the descending branch of the plasticity limit criterion. When the stressesare decreasing, the material dilatancy angle should decrease. If it reaches the imposed volumevariation rate, an asymptotic stress state is met, and failure will develop; strains still evolve, whereasthe stresses remain constant. If the dilatancy angle does not reach the imposed rate before thestresses vanish, liquefaction will develop (see reference [36] for a general liquefaction criterion). Onthe contrary, if the imposed volume variation rate is lower (i.e., more contractant or less dilatant)than the material volume variations rate, stresses increase along or close to the ascending branch ofthe Mohr–Coulomb criterion. The dilatancy angle is decreasing along with this stress increase.When the related volume variation rate is equal to the imposed volume variation rate, an asymptoticstress point is again reached, accompanied by failure.

We use this consideration to determine the value of c for the INL2 case. Different values of R arechosen, and for each value, different initial stress states are considered. Examples of the results of thetest for du/dg =� tan(10�) appear in Figure 20. The curves of Figure 20 show that, regardless of theinitial state, the stresses reach the Mohr–Coulomb criterion during this slightly dilatant shearingpath. Then, after some additional changes (increases or decreases, depending on the previousdiscussion) in stresses on the criterion, a final stress point is reached: t and s do not evolve anyfurther. As explained above, this case occurs when the loading parameters correspond to the flowrule of the INL2 relation (its existence is also proved by these results). For other values of R,another final stress point, still on Mohr–Coulomb criterion but with a different value of s, isreached; as for Quadri case, the dependence of the dilatancy angle on s is obtained here in INL2case. For example, the results in Figure 20 show that 10� =c(s� 3.3 MPa). The collected values ofc(s) for the INL2 relation will finally be plotted in Figure 22 (in section 6.4).

As the values of c(s) for the INL2 and Quadri relations are not exactly equal (a comparison appearsin Figure 22), predictions of these two relations show some differences near the plastic limit criterionalong these proportional displacement loading paths (see Figure 21). With both relations, four paths areimposed for the same control parameter, R=� tan(10�), from two different initial stress states. Asexplained earlier, all the curves converge on the stress states of the Mohr–Coulomb plastic criterion,for s values such that c(s) = 10�. The differences in the c(s) values for both relations lead here todifferent behaviours on plastic criterion. In this case, we obtain on the criterion ds< 0 for theQuadri-performed paths (a contractant behaviour) and ds> 0 for the INL2-performed paths (adilatant behaviour), whereas the differences are smaller away from the criterion.


0 5 10 15 200

5

10

15

γ (mm)

σ (M

Pa)

σ0=5 MPa

σ0=10 MPa

σ0=15 MPa

0 5 10 15 200

0.5

1

1.5

2

2.5

γ (mm)

τ (M

Pa)

0 5 10 150

0.5

1

1.5

2

2.5

σ (MPa)

τ (M

Pa)

M−C

σ0=5 MPa

σ0=10 MPa

σ0=15 MPa

σ0=5 MPa

σ0=10 MPa

σ0=15 MPa

Figure 20. Stress evolutions for proportional displacement loading with du/dg=� tan(10�), from three initialstress states (INL2 constitutive relation).


6.4. Flow rule for discrete element method simulations

Reference [2] discussed the plasticity limit criterion of the used discrete numerical model for infilledrock joints, without focusing on flow rule. Nevertheless, the discrete results also allow to point outthis feature, in at least two ways.

0 5 100

1

2

σ (MPa)

τ (M

Pa)

σ0=5 MPa INL2

σ0=10 MPa INL2

σ0=5 MPa Quadri

σ0=10 MPa Quadri

Figure 21. Proportional displacement loading paths (du/dg =� tan(10�)) for Quadri and INL2 relations, fromtwo different initial stress states. For each relation, both curves reach the same final state for which flow rule

corresponds to loading parameter: c(s) = 10�.



First, the final slopes of the u(g) curves for the CNL tests (Figure 8), once on the plateau of t, aremeasured, and the corresponding values of c are deduced. They are plotted according to s inFigure 22. Errors (which are caused by nonlinearity of discrete u(g) curves in Figure 8(b)) for thesevalues are plotted as error bars. For the two values corresponding to s=10 or 20MPa, the bars are infact contained in the marker that is used. Figure 22 gathers the evolutions of c with s according tothe three different models used: the Quadri and INL2 relations and the discrete numerical model. Allthe models present a dilatancy angle that decreases with s, and the values of c for the differentmodels are close. The main point is that the dilatancy angle is clearly different from the frictionangle (which is here equal to 29): the rock joint behaviour, approached from the three differentperspectives, is nonassociated (Figure 22).

Second, other discrete results indicate the flow rule. Similar to what was done with INL2 relation, differentproportional displacement loading paths are performed with the discrete model. The tests are governed byparameter θdl; we impose the condition du� dgtan(θdl) = 0 (R= tan(θdl)). All the tests are performed fromthe mechanical state reached at the end of a CNL test under s=10MPa. The corresponding changes instress appear in Figure 23. Figure 23(a) shows that the tests for θdl2 [� 5� ; 15�] allow the stresses toincrease, whereas the tests with θdl2 [� 20� ;� 5�] allow the stresses to decrease. As Figure 23(b) clearlyshows, we can assume that for tan(� 10�)< du/dg< tan(� 5�), the stresses would not evolve and that theflow rule would be obtained. This result confirms the previous discrete results, which showed already that5� <c(s=10MPa)< 10�.

The discrete results in Figure 23(a) also can be compared with those in Figures 3 and 14. As in theQuadri and INL2 cases, the loading paths from the Mohr–Coulomb criterion with du> 0 move therock joint stress state further inside the criterion; whereas the paths with du< 0 allow the stress stateto remain along the criterion (even if stresses evolve).

This flow rule determination also allows commenting on the horizontal stress thresholds (for theDEM results) of Figure 1. As evocated in paragraph 3.1, our relations did not recover these stressplateaus. It should now be clear that such stress plateaus should occur when the loading parameterscorrespond to the material flow rule. However, certain findings suggest that this did not occur forthe DEM results of Figure 1. First, the final stress states for these CND discrete simulations do notcorrespond to the plastic limit criterion deduced from other discrete results (such as the CNL tests),as the maximum ratio t/s is in fact lower for these CND discrete simulations than for the CNLdiscrete simulations [2]. Second, the discrete CND stress levels would correspond to a discrete flowrule with dilatancy angle equal to zero whatever s (see Figures 4 and 5, in which the levels arereached for different stress states). However, this result contradicts two other sets of discrete resultsthat reveal a non-null dilatancy angle changing with respect to s. Hence, we interpret these

0 5 10 15 20 25 300

5

10

15

20

σ (MPa)

ψ (d

egre

es)

QuadriINL2DEM

Figure 22. Evolution of c depending on s, a comparison between Quadri and INL2 constitutive relations,and DEM results.


0 5 10 15 200

2

4

6

8

10

σ (MPa)

τ (M

Pa)

θdl=15

θdl=10

θdl=5

θdl=0

θdl=−5

θdl=−10

θdl=−15

θdl=−20

Initialstate

(a) In Mohr plane

6 6.5 7 7.5 8 8.54

6

8

10

12

γ (mm)

Str

esse

s (M

Pa)

θdl=−10 : τθdl=−5 : τθdl=−10 : σθdl=−5 : σ

(b) According to γ, for loading parameters aroundthe flow rule

Figure 23. Stress changes for different proportional displacement loading paths (with du/dg = tan(θdl)), froma completely sheared state.


thresholds as a bias induced by localisation, once excessive values for g have been reached at the end ofcertain simulations that use the numerical model.

6.5. (Non)-Associativity of flow rule

Figure 22 shows that our two rock joint relations lead to nonassociated behaviour. In the elasto-plasticframework, with direct definitions of the plastic limit criterion f and plastic potential g, it is known thatnonassociativity corresponds to nonsymmetric constitutive matrices. As our framework is different (nodirect definitions of f and g, even if such features finally appear), we investigate this possible linkbetween the symmetry of the constitutive matrix and associativity for both the Quadri and INL2relations.

On the plastic limit criterion, cancellation of the determinants of Quadri matrices is respected forboth matrices of the tensorial zones implying du< 0 (see paragraph 4 and Figure 12). Moreover, forany increment (dg, du< 0), Figure 3 shows that stress response ds to this (dg, du< 0) correspond

either to limit stress state (!ds ¼ !

0, when (dg, du) corresponds to flow rule) or to a neutral loading.Neutral loading means that the rock joint stress state stays during the entire loading on the plasticlimit criterion (Mohr–Coulomb): t = stan(’) and dt= dstan(’). Considering such neutral loadings,and starting from the cancellation of the determinant of corresponding constitutive matrices,Equation (9), we obtain:

det Mð Þ ¼ 0 , Gg ¼ Gu

NuNg⇒Ggdg ¼ Gu

NuNgdg by multiplying by dg

, Guduþ Ggdg ¼ Gu

NuNuduþ Gu

NuNgdg by addingGudu

(16)

Expressions for dt and ds finally appear (see, e.g., Equation (8)), and we find that, on the plasticlimit criterion

dt ¼ Gu

Nuds , Gu

Nu¼ dt

ds¼ tan ’ð Þ (17)

Equation (17) is valid for any piecewise linear rock joint relation with constitutive matrices whosedeterminants vanish along such neutral loadings. Obviously, the chosen expressions for theG�

u andN�u



moduli of Equation (11) obey Equation (17) (as explained previously, the du> 0 tensorial zones, thenthe Gþ

u and Nþu moduli, are not to be considered).

If the matrix is symmetric, we have Ng =Gu, and it appears that the previous Equation (17)corresponds to the flow rule definition of Equation (14), with c=’. Symmetry of the constitutivematrix is here also equivalent to an associated behaviour.

To verify this result numerically, for the INL2 relation as well, new expressions for some of moduliare proposed in Equation (18). All other expressions remain as in Equations (10) and (11). In this case,the symmetry of the M+� matrix of Table I, which corresponds to dg> 0 and du< 0, is obtained (withthe exception of the contractant behaviour), along with cancellation of its determinant on plasticitylimit criterion.

Gþg ¼ G0

g � G0g � Gf

g

� � 1� e�

t=stan ’ð Þ

1� e�1

0BBB@

1CCCA

ss0

� �13whereGf

g ¼ NdNfutan ’ð Þ2

G�u ¼ NdNf

utan ’ð Þ ss0

� �13

Nþg ¼ G�

u if t=s > tan ’cð Þ

(18)

Different simulations are then performed using these new ‘symmetric’ Quadri and INL2 relations todetermine if an associated behaviour is obtained numerically in these cases.

First, two CNL shearing paths, for s= 1 and 20MPa, are simulated with both these relations (seeFigure 24). Slopes of u(g) curves can be observed in Figure 24(a). Once the t-plateau is reached(which occurs in all cases for g⩾2:5 mm, see Figure 24(b)), the final slopes do not depend anymoreon the s values—the curves are parallel—and c=’= 29� is obtained precisely for the Quadri caseand approximately (with slight differences) for the INL2 case. As experiments show that thedilatancy rate of rock joints decreases with s [6,27,37], such results illustrate the need to considernonassociativity when simulating their behaviour, as it has been previously discussed by Plesha [8]for example.

Second, proportional loading paths are used with two different imposed volume variation rates:R= tan(25�)< tan(’), and R = tan(35�)> tan(’) (Figure 25). Consistent with the previous discussionin paragraph 6.3 and with equality c=’, an ‘infinite’ increase in stress values is observed forR= tan(25)> tan(’), whereas the stresses end to vanish for R= tan(35�)< tan(’).

0 1 2 3 4 5 6−4

−3

−2

−1

0

1

γ (mm)

u (m

m)

INL2Quadri

σ=1 MPa

σ=20 MPa

(a) (γ)

0 1 2 3 4 5 60

2

4

6

8

10

12

γ (mm)

τ (M

Pa)

INL2Quadri

(b) (γ) for σ =20 MPa

Figure 24. Results of CNL shearing paths (s =1 or 20MPa) for symmetric cases of both relations.


0 10 20 30 400

5

10

15

20

25

σ (MPa)

τ (M

Pa)

INL2Quadri

R=tan(25o)

R=tan(35o)

(a) Mohr plane

0 2 4 6 80

20

40

60

80

γ (mm)

σ (M

Pa)

INL2Quadri

R=tan(25o)

R=tan(35o)

(b) σ (γ)

Figure 25. Proportional displacement loading paths simulated with both relations, from (t= 3MPa ; s = 10MPa), for imposed volume variation rates R around tan(c) (c =’).


6.6. Singularity or regularity of the plastic potential

A more general discussion on the flow rule question can be found in reference [35], including studieswith soils using the INL2 and Octo relations. One issue raised by the corresponding authors, Darve andNicot, was the singularity of the flow rule (the plastic potential, strictly). As the constitutive relationsused present different tensorial zones, different constitutive matrices exist for a given stress state,depending on the loading direction. For this stress state, it would thus be possible to obtain differentplastic strain directions corresponding to different flow rules, depending on the different matrices.This situation would lead to a singular plastic potential with vertex effects, also called ‘cornertheories in elasto-plasticity’. Such cases occur for general 3D conditions. Although it has beenshown that, for rate-independent (i.e., elasto-plastic) materials under 2-dimensional (2D) conditions(e.g., axisymmetry or plane stress), a regular flow rule exists at failure as the eigenvector related tothe first vanishing eigenvalue of the elasto-plastic matrix, when the plastic limit condition is reachedwith a vanishing value for the determinant of the elasto-plastic matrix. We note that sucheigenvectors corresponding to the flow rule were presented in Figure 13, for our case.

Indeed, as the framework has been stated as 2D in this paper, it has been shown that a regular flow

rule can be exhibited. As we have shown, the loading direction!dl leading to

!ds ¼ !

0 can finally belongto only one tensorial zone. It is remarkable that in all of the previous discussions, this regularnonassociated flow rule was obtained both by two constitutive relations and by direct discretesimulations.

7. CONCLUSION

Constitutive modelling of rock joints has been discussed. To obtain powerful constitutive relations,relations with a high number of tensorial zones have been used. A quadrilinear relation (Quadri,with four tensorial zones) and an incrementally non-linear, second-order (INL2, with infinitetensorial zones) relation were then proposed. These two relations correspond to what was performedfor soils [3,17] with an octolinear (with eight tensorial zones) and also an INL2 relation. The INL2and Quadri relations were compared throughout the paper to illustrate the differences between thesetwo types of relations, which correspond to two different interpolations. From several perspectives,such differences were generally small (with the exception of the case shown in Figure 21).Nevertheless, the INL2 relation seems to be a little more accurate, whereas the Quadri relation offersthe possibility to lead analytical developments. Our relations were calibrated using a numericaldatabase from a discrete model of an infilled rock joint [2]. In addition to a high (even infinite, in



the case of INL2) number of tensorial zones, the relations have the advantages of linking normal andtangential variables and of considering the dilatancy or the effect of the compressions on t variations,depending on the shearing state of the joint. The predictive abilities of the relations also werediscussed.

The relations are defined outside the classical elasto-plastic formalism. However, during calibration,proper care was taken to ensure that the proposed constitutive relations respect the theoreticalrequirements. It was analytically proven and numerically verified that both the constitutive relationsfulfill the continuity condition (or the so-called ‘consistency condition’ in classical elasto-plastictheory). The relations also reflect the existences of a plasticity limit criterion and a flow rule,without introducing these features directly. Indeed, analytical and numerical analyses were presentedand showed that both our relations obey a Mohr–Coulomb plasticity limit criterion (withoutrealising any correction for stress values). For the flow rule, the values of dilatancy angle weredetermined for the three different models presented here (Quadri or INL2 relations, or DEM), whichwere based on infilled rock joints. All the results reveal similar values for the dilatancy angles thatdecrease according to s. It is important to note that our models correspond to nonassociatedbehaviours. We analytically and numerically illustrated that associativity is obtained for symmetricconstitutive matrices, as in classical elasto-plasticity. Associativity should be avoided in rock jointsimulations because of the observed decreasing dilatancy with s, for example.

Because of this nonassociativity, different kinds of rock joint failures may occur: especially on orbefore the plasticity limit criterion [38–40]. A material stability analysis will thus be led, by use ofthe ‘second-order work criterion’ [25,41,42], which is the only one available for predicting all kindsof failures if flutter instabilities are excluded. Using these two tools, the INL2 constitutive relation(which has been implemented in UDEC [43]) and the second-order work criterion should allow acomprehensive mechanical analysis of rock slope stability.

REFERENCES

1. Bandis SC, Lumsden AC, Barton NR. Fundamentals of rock joint deformation. International Journal of RockMechanics and Mining Science and Geomechanics Abstracts 1983; 20(6):249–268.

2. Duriez J, Darve F, Donze FV. A discrete modeling-based constitutive relation for infilled rock joints. InternationalJournal of Rock Mechanics and Mining Sciences 2011; 48(3):458–468. doi:doi:10.1016/j.ijrmms.2010.09.008.

3. Darve F, Labanieh S. Incremental constitutive law for sands and clays. simulations of monotonic and cyclic tests.International Journal for Numerical and Analytical Methods in Geomechanics 1982; 6:243–275.

4. Darve F. The expression of rheological laws in incremental form and the main classes of constitutive equations. InGeomaterials Constitutive Equations and Modelling, Darve F (ed.). Elsevier Applied Science: London, 1990; 123–148.

5. Barton N, Bandis S, Bakhtar K. Strength, deformation and conductivity coupling of rock joints. InternationalJournal of Rock Mechanics and Mining Science and Geomechanics Abstracts 1985; 22:121–140.

6. Leichnitz W. Mechanical properties of rock joints. International Journal of Rock Mechanics and Mining Science andGeomechanics Abstracts 1985; 22(5):313–321.

7. Saeb S, Amadei B. Modelling rock joints under shear and normal loading. International Journal of Rock Mechanicsand Mining Science and Geomechanics Abstracts 1992; 29(3):267–278.

8. Plesha ME. Constitutive models for rock discontinuities with dilatancy and surface degradation. InternationalJournal for Numerical and Analytical Methods in Geomechanics 1987; 11:345–362.

9. Gens A, Carol I, Alonso EE. A constitutive model for rock joints formulation and numerical implementation.Computers and Geotechnics 1990; 9:3–20.

10. Wang J, Ichikawab Y, Leung C. A constitutive model for rock interfaces and joints. International Journal of RockMechanics and Mining Sciences 2003; 40:41–53.

11. Souley M, Homand F, Amadei B. An extension to the saeb and amadei constitutive model for rock joints to includecyclic loading paths. International Journal of Rock Mechanics and Mining Science and Geomechanics Abstracts1995; 32:101–109.

12. Hill R. Continuum micro-mechanics of elastoplastic polycristals. Journal of the Mechanics and Physics of Solids1965; 13:89–101.

13. Hill R. Generalized relations for incremental deformation of metal crystals by multislip. Journal of the Mechanicsand Physics of Solids 1966; 14(2):95–102.

14. Hill R. The essential structure of constitutive laws for metal composites and polycrystals. Journal of the Mechanicsand Physics of Solids 1967; 15(2):79–95.

15. Saada AS, Bianchini G (eds.). Constitutive equations for granular soils, Balkema: Rotterdam, 1987.16. Nicot F, Darve F. Basic features of plastic strains: From micro-mechanics to incrementally nonlinear models.

International Journal of Plasticity 2007; 23:1555–1588.



17. Darve F, Flavigny E, Méghachou M. Yield surfaces and principle of superposition revisited by incrementallynon-linear constitutive relations. International Journal of Plasticity 1995; 11(8):927–948.

18. Darve F, Nicot F. On incremental non-linearity in granular media : phenomenological and multi-scale views (Part I).International Journal for Numerical and Analytical Methods in Geomechanics 2005; 29:1387–1409.

19. Nicot F, Lambert C, Darve F. A new constitutive relation for rock joints calibrated by means of a discrete elementmethod. In Deformation Characteristics of Geomaterials, DiBenedetto H, Doanh T, Sauzéat C (eds.). Swets &Zeitlinger: Lisse, 2003; 1257–1262.

20. Royis P, Doanh T. Theoretical analysis of strain response envelopes using incrementally non-linear constitutiveequations. International Journal for Numerical and Analytical Methods in Geomechanics 1998; 22:97–132.

21. Cundall P, Strack O. A discrete numerical model for granular assemblies. Geotechnique 1979; 29:47–65.22. Kutter HK, Raunterberg A. The residual shear strength of filled joints in rock. Proc. 4th Symp. ISRM, 1979; 221–227.23. Papaliangas T, Hencher SR, Lumsden AC, Manolopoulou S. The effect of frictional fill thickness on the shear

strength of rock discontinuities. International Journal of Rock Mechanics and Mining Science and GeomechanicsAbstracts 1993; 30:81–91.

24. Pereira JP. Rolling friction and shear behaviour of rock discontinuities filled with sand. International Journal of RockMechanics and Mining Sciences 1997; 34:244.e1–244.e17.

25. Darve F, Servant G, Laouafa F, Khoa H. Failure in geomaterials : continuous and discrete analyses. ComputerMethods in Applied Mechanics and Engineering 2004; 193:3057–3085.

26. Stephansson O, Jing L. Testing and modeling of rock joints. In Proc. of the MJFR-2 Conf., Rossmanith HP(ed.).Balkema: Rotterdam, 1995.

27. Boulon M, Armand G, Hoteit N, Divoux P. Experimental investigations and modelling of shearing of calcite healeddiscontinuities of granodiorite under typical stresses. Engineering Geology 2002; 64:117–133.

28. Smilauer V, Catalano E, Chareyre B, Dorofeenko S, Duriez J, Gladky A, Kozicki J, Modenese C, Scholtès L, SibilleL, et al. Yade Documentation. 1st edn., The Yade Project, 2010. http://yade-dem.org/doc/.

29. Radjai F, Wolf DE, Jean M, Moreau JJ. Bimodal character of stress transmission in granular packings. PhysicalReview Letters 1998; 80(1):61–64.

30. Cambou B, Jean M. Micromécanique des milieux granulaires. Hermes Science: Paris, 2001.31. Calvetti F, Viggiani G, Tamagnini C. A numerical investigation of the incremental behavior of granular soils. Rivista

Italiana di Geotecnica 2003; 3:11–29.32. Sibille L, Donzé FV, Nicot F, Chareyre B, Darve F. From bifurcation to failure in a granular material: a dem analysis.

Acta Geotechnica 2008; 3(1):15–24.33. Donzé FV, Richefeu V, Magnier SA. Advances in discrete element method applied to soil, rock and concrete

mechanics. Electronic Journal of Geotechnical Engineering 2009; Special Volume, Bouquet 08:1–44.34. Gudehus G. A comparison of some constitutive laws for soils under radially symmetric loading and unloading. Proc.

3 rd Numer. Meth. in Geomechanics, 1979.35. Darve F, Nicot F. On flow rule in granular media : phenomenological and multi-scale views (part ii). International

Journal for Numerical and Analytical Methods in Geomechanics 2005; 29:1411–1432.36. Darve F. Liquefaction phenomenon of granular materials and constitutive instability. International Journal of

Computer Engineering 2005; 13:5–28.37. Ladanyi B, Archambault G. Simulation of shear behavior of a jointed rock mass. Proceedings of the11th Symposium

on Rock Mechanics 1970; 1:105–125.38. Desrues J, Chambon R. Shear band analysis and shear moduli calibration. International Journal of Solids and

Structures 2002; 39(13–14):3757–3776.39. Darve F, Sibille L, Daouadji A, Nicot F. Bifurcations in granular media: macro- and micro-mechanics approaches.

Comptes Rendus Mécanique 2007; 335:496–515.40. Khoa H, Georgopoulos I, Darve F, Laouafa F. Diffuse failure in geomaterials: Experiments and modelling. Computers

and Geotechnics 2006; 33:1–14.41. Sibille L, Nicot F, Donzé FV, Darve F. Material instability in granular assemblies from fundamentally different

models. International Journal for Numerical and Analytical Methods in Geomechanics 2007; 31:457–481.42. Nicot F, Darve F. Diffuse and localized failure modes: Two competing mechanisms. International Journal for

Numerical and Analytical Methods in Geomechanics 2011; 35(5):586–601. doi:10.1002/nag.912.43. Itasca. UDEC Version 4.0: Use Manual, 2007.


http://yade-dem.org/doc/

Incrementally non-linear plasticity applied to rock joint modelling

Documents

Transcript of Incrementally non-linear plasticity applied to rock joint modelling