You are on page 1of 20

INTERNATIONAL JOURNAL FOR NUMERICAL AND ANALYTICAL METHODS IN GEOMECHANICS. VOI. 6.

109-128 (1982)

LIMIT ANALYSIS: A LINEAR FORMULATION OF THE


KINEMATIC APPROACH FOR AXISYMMETRIC
MECHANIC PROBLEMS
S. TURGEMAN AND J. PASTOR
Laboratoire de Micanique des Sols de Grenoble, B.P. 53 38041 Grenoble Cedex, France

SUMMARY
This paper deals with a numerical formulation of the kinematic limit analysis method. This formulation,
based on a finite element discretization, is available for axisymmetric cases. The linearization of the
yield Tresca and Von Mises criteria leads to linear programming problems. The velocity fields described
here are plastically admissible only on an average. The influence of this approximation is quantified for
the triaxial test problem and the stability of circular excavations.

1. INTRODUCTION
The kinematic and static limit analysis methods allow us to determine an upper and a lower
bound respectively for the load of a rigid-perfectly plastic system obeying the maximum work
principle.
However, the analytical application of these methods (especially the static method), even
in such a favourable case as plane strain, is arduous and the results, once obtained, are not
always sufficiently accurate.'"
This reason explains why it was felt necessary to resort to numerical models involving
discrete unknown fields in finite elements. Fremond and Salencon' have developed a plane
strain kinematical program for the standard Von Mises material. Mercier4 has invested a
visco-plastic numerical approach of the rigid-perfectly plastic model. The authors6 have
developed a plane strain kinematical formulation leading to a linear programming problem
in the case of Von Mises and Coulomb standard materials.
The aim of the present study is to extend the preceding technique to axisymmetric problems
in the case of a standard Von Mises or Tresca material. After recalling the basic principles
of the kinematic method, the axisymmetric formulation is related in detail in the case of the
Von Mises material. We furnish later the modification involved by to Tresca material study.

2. THE KINEMATIC METHOD'


(A) denotes a system of rigid-perfectly plastic solids, obeying the maximum work principle*
and submitted to a loading process. u denotes a statically admissible (S.A.) stress field and u

* The behaviour of a material obeying the maximum work principle is defined by a convex plasticity criterion and
the normality Row law. A standard material has a normal flow law.

0363-9061/82/010109-20$02.00 Received 24 June 1980


@ 1982 by John Wiley & Sons, Ltd. Revised 2 December 1980
110 S. TURGEMAN AND J. PASTOR

a kinematically admissible (K.A.) strain rate field. If two linear operators Q and 4:
Q :u S.A. + Q (u)= (Q1, . . . , Q,, ) E R"
.. ,~,,)ER"
4 : K.A.+cj(~)=(ql,.
~
allow us to write the virtual energy rate theorem:

1"~. u d V = Q(a). q ( v ) ( V :volume of (A))

then (A) is subjected to a n parameters (Q1,. . . , Q,,) loading process.


Any admissible stress field (+ (S.A. and plastically admissible (P.A.)) defines an admissible
loading Q(u).
Let K = {Q(a),a admissible} be the convex of the admissible loadings for (A). Every limit
load (corresponding to the collapse of (A)) belongs to the boundary F(K)of K.
The kinematic theorem can be formulated as: any admissible strain rate field u (K.A. and
P.A.) defines an external or tangent to the K hyperplane with the equation 0 4 ( u ) = P ( u ) ,
P ( u ) being the total plastic dissipated power for the given strain rate field I).An external
approach of F(K)is obtained through the kinematic method. When solving the following
minimization problem:
P ( 6 )= Min ( P ( u ) , u admissible, 4 ( u ) =q d )
for one direction 4d of W", the hyperplane H of equation: Q. g d = P ( 6 ) is tangent to K
(Figure 1).

dK)
Figure 1. Kinematic method

3. LINEAR FORMULATION OF THE KINEMATIC METHOD


FOR THE VON MISES MATERIAL
In the euclidian space R3, we consider a cylindrical coordinate system r, 6, z.
Since the flow is assumed to be meridian, the velocity field can be written as:
ur = u,(r, I); uz = u,(r, I); ug = 0

The strain rate fields u are expressed in terms of the velocity displacements:
KINEMATIC APPROACH FOR AXISYMMETRIC PROBLEMS 111

Any section of the system along the Oz symmetry axis is divided into two complementary
zones 2, and 2, (Figure 2). The 2, discretized zone contains the assumed strained area. The
2,zone is rigid. Strain rate fields are obtained in 2, by means of a finite element discretization.
These fields are defined in 2, by the boundary conditions.

0 r

Figure 2. Finite element mesh

The 2, zone is subdivided into triangular finite elements T of nodes Nk.The displacement
velocity fields u vary linearly on each T, and are constant in 2,.Discontinuity lines may occur
between adjacent elements along Lj sides of the triangles. These fields are defined only by
their nodal components (urk,uzk) at Nk. Considering a Titriangle, for instance, we have:

ur(r,z ) = a + fir + yz
uz(r, z ) = a + P r + yz

a,0, y (respectively a,
p, y ) are linear functions of Urk (u,t respectively). A classical derivation
of u, and uz gives us:

For this field D, a linear formulation of the dissipated power can be obtained by hearing
the Von Mises criterion.

3.1. Linearization of the Von Mises plasticity criterion


The Von Mises yield surface (SM)is included in the four-dimensional vector subspace E of
the R9 stress components u,.,uh u,,r,,, r,, r,& rez, rrh rep The E space is defined by the
linear equations:

rr2= T,,; r z e = r e z = r m = rer =O (5)

(the material being isotropic, u and u have the same principal directions) the explicit equation
of SM is:
2 2 2
f ( u ) = ( U e - u , ) +(u,-u,) + ( U , - U ~ ) ~ + ~ T =
? O~ - ~ C (6)

A linearization is readily obtained from the formulation of SM in the vector space R4defined
by the X, Y,2, T directions. We get u = (u,. . ., ror)in R9 from S = (X, Y,2, T) in R4 from
112 S. TURGEMAN AND J. PASTOR

u,= ( - 2 Y + J ( ~ ) z )J6
/
a; = ( - J ( 3 ) X + Y + J ( ~ ) z )J6
/
L ( S )= 0- =< (7)
T,, = T/d2

r,, = T/J2

LTr@= 7 8 , = r e z = T,e = 0

F ( S )= x 2+ Y 2+ T 2- 2 2 = 0 (8)
This is the equation of a cylinder HC with axis OZ.
A polyhedral approach of the X 2 +Y 2 +T 2= 2c2 sphere in R3 is obtained by the planes
gi(X, Y, T )= ~ , X X + ~ Y + C ~ T =- JO (( ~ i =) 1C, . . . ,f i ) with a t + b f + c ; = 1 (Figure 3 ) . The
H P polyhedral cylinder defined by the hyperplanes of the equation: G i ( X ,Y,2, T )=
g,(X, Y, T )= O ( i = 1, . . . , 61)in R4 is an external approach of HC. The yield surface of the
linearized Von Mises criterion (L.M. criterion) is HP.

3.2. Dissipated power in a P.A. v field for the L.M. criterion


In R4, the components vx, uy, uz, U r of a u field, in a standard material obeying the L.M.
criterion defined above, are given by:

or similarly:

The hi ( i = 1 , . . . , +i) are real values with the conditions:


AisO if Gi(X,Y , Z , T ) = O
Ai =0 if G i ( X ,Y, 2, T )< 0
The dissipated-power function is:
A
l I L ~ ( v=) 1 J ( 2 ) c A i ( A i according to (12))
i=l
KINEMATIC APPROACH FOR AXISYMMETRIC PROBLEMS 113

X i

Figure 3. Linearization of the X 2 + Y 2+ T 2= 2c2 sphere (61


= 26)

or:

(11)is the necessary and sufficient condition to the conclusion: 'P.A. u field'. This is the same
condition as for a standard Von Mises material. Any u field P.A. for the L.M. criterion is
P.A. for the Von Mises criterion. The corresponding dissipated power for the L.M. criterion
is more important than the dissipated power in u for the Von Mises criterion.
O n a 17; triangle, equation (11) is equivalent to the 3 following linear equations:
a (Urk = 0 ; 2P ( u r k + y r ( U z k ) = 0; y ( u r k ) =0 (15)
According to equation (1l), the solutions u must be written as:

u, = a' + P'r --z


P
ur = pr ;
2
with:
/3 a constant in the zones of Z , where u is continuous.
They constitute an empty or trivial set (except for completely discontinuous meshes).
The condition:
tr ( u ( G i ) ) = ( u r + v , + u , ) ( G i ) = O (Gi: centreof massof Z) (16)
is substituted for efficiency to the too strong condition (11). According to (16), the u fields
are P.A. only on an average (see in Section 4 the accuracy related to this approximation).
114 S. TURGEMAN A N D J. PASTOR

For any triangle z, 6 positive variables A t related to the 6 nodal variables urk and uzk are
defined in (17):
cPr(A = p(urk)
Pz (A;) = Y ' ( U z k ) (17)
Prz (A i ) = (Y(urk)+@'(uzk))/2
which results from (4) and (10).
The dissipated power in u, defined on T, by the variables u r k and &k which are solutions of
equations (16) and (17) ((16) is a linear equation of u,k, uzk) can be expressed, according to
(14),in the following form:

with: STi:area of the triangle z, generating the volume VK,rGi: radial coordinate of Gi.
3.3. Dissipated power in the discontinuity lines
The considered velocity fields may involve discontinuities. It is necessary to determine the
condition upon which these fields are P.A. along discontinuities.
The discontinuity surfaces are surfaces of revolution SLi determined by the rotation of the
discontinuity sides ( L i ( M : M
, : ) around Oz. Let n and t be respectively the normal and tangent
unit vector to Li (Figure 4).

12

Figure 4. Discontinuity side Li

The projections of the discontinuity velocity vector at point M are [u,] on n and [u,] on t
respectively. Plastically admissible velocity fields for the Von Mises criterion are obtained when:
V M E Li: [U"](M)= 0 (19)
This relationship is equivalent to the two linear equations in terms of u r k , uzk:
) 0;
[ u , ] ( ~ j= j = 1,2 (20)
Then, the corresponding dissipated power in SLi is:
KINEMATIC APPROACH FOR AXISYMMETRIC PROBLEMS 115

Assuming a value to si such that: si[uI] = I[u,]l, the dissipated power can also be put into a
linear form:

3.4. Conclusion
The dissipated power has been calculated in the strain rate field u defined by nodal variables
Urk and U , k :
on Ti triangles: linearly for the h j ( A i a 0 ) by substitution of the LM criterion for the Von
Mises criterion;
along the discontinuity areas: linearly for the nodal variables Urk and uzk,the sign of the
discontinuities being fixed.
The direction of the discontinuities can be readily fixed in the case of local discontinuities.
The relationships (17)between A { and u r k , u z k and the u P.A. condition are linear (remember
that u is P.A. only on an average in a T, triangle).The u field must be K.A. The u K.A.
relationships are carried into linear expressions if the boundary conditions are linear or
piecewise linear. In the same way, the equations 4 ( u ) = ( i d and U, = O on Oz are linear
expressions of Urk, u z k .
The complete set of relationships is written as:

where [A] denotes a matrix, { B } a vector and {X} the total vector of variables urk, u z k , A:.
The kinematic method leads to a linear optimization form as:

Let ij be the optimal strain rate field obtained from the solution of (9).The Von Mises
dissipated power P*(6)can then be calculated:

P*( 6 ) = c J(26jj6ii)d V + 1Ps,,( 6 )

where V, denotes the volume of the body of revolution associated with 2,.
The hyperplane P*(C)-Qq,=O defines an approach of F ( K ) which is not necessarily
external to K,the 6 field being P.A. only on an average.

4. TESTS
The purpose of the following tests is to quantify the approximation u P.A. on an average
on the result obtained from (9)(see Fremond et a1. for a mechanical interpretation of the
dualization of conditions u P.A. in the strained areas).
116 S. TURGEMAN AND 1. PASTOR

4.1. Absolute incertitude due to the approximation uPAon an average


Let MTb be a Von Mises material with a restricted P.A. stress domain, defined by the
following conditions:
(we- c,)+ (nr-a,) + (u2- g o ) + 6 2 2
, ~ 0
~ - 6~
{ ( ko
Itr (u)l=lu8+u,+ u 2s ( k o :positive value)
(24)

(A) is a standard Von Mises system with a cohesion c. For simplication, (A) is assumed to
be submitted to a loading process with a single loading parameter Q. M,r is a mesh correspond-
ing to the limited area 2,. D is the set of the u fields in Mer which are:
K.A.;
P.A. in the discontinuities;
P.A. only on an average in the strained zones.
The best field of D is 5 ; it is obtained when solving the (9) problem:
P ( 6 )= min ( P ( u ) ,u E D )
Let us recall that the loading Q = P * ( 6 ) / q dis not required to be external to the K convex
of the admissible loadings of (A). We try to find a real positive A(6) value such that:
Q = (P*(6)+ A(6))/(id is necessarily external to K.
Let u* be an optimal admissible stress field (i.e. Q ( u * )is the limit load). Itr (a*)[
is limited
in 2, by a positive real value ko. The (A) composed of the MTbstandard material (see (24))
inside 2, and the Von Mises standard material outside the 2, region is such that u* is also
an optimal admissible stress field. As a consequence the formulation of the kinematic method
for the MTk, standard material leads to an external approach of K . Any u field is P.A. for
the MTk, standard material. Pk,(u) is the corresponding dissipated power in such a u field.
So, we get:
Q =pk,(u)/qd: external to K, for any u K.A.

Particularly, the 6 field leads to Q = Pk,,(6)/& outside K.


The relationship

with

leads to the conclusion that:


Q = ( P * ( 6 ) + 775 ~ ( 6 ) ) / 4 d isoutside K

The value of A($) is: k o ~ ( u ^ ) / when


3, an optimal admissible stress field u* is known. A($
is the absolute incertitude on the P*(C)value, due to the condition u P.A. on an average only.
Generally, the optimal admissible stress fields are unknown. So, ko and A(;) cannot be
calculated. Concerning the example investigated in the present study, the known admissible
stress fields, certainly not optimal, allowed us to obtain an acceptable approximation of ko.
Then, it is possible to fix approximately the error involved by the computation resulting from
the average plastic condition on the u fields.
KINEMATIC APPROACH FOR AXISYMMETRIC PROBLEMS 117

4.2. Ultimate compressive strength on a cylindrical sample


A cylindrical Von Mises sample (R = 2 H ) (Figure 5 ) is considered here. The top is submitted
to a downward translation at a constant vertical velocity Uo. The bottom, DC, stands on a
fixed support. The interfaces AB and C D are frictional standard interfaces (also called perfectly
rough standard interfaces) (see, Reference 5, pp. 146-149). In this case, the set of P.A.
b 0; 17)s c(un,T : normal and tangential stresses) and
stresses at the interface is defined by: a,,
the normal flow rule implies: [ U,,]= U,,,- U,,,S 0 (Figure 5 ) .

Figure 5. Ultimate compressive strength on a cylindrical sample

The boundary conditions for the complete system in the meridian plane Orz are:
on AB: U, = U O ;u , = O ; ue=O;
on DC: U, = U, = U e = 0 ;
on A'D' and on B'C': ar= T,, = T , =
~ 0.
The problem involves a loading parameter

Q= (Is azdS)/cS

and the associated strain rate parameter is (id= cSUo. Kobayashi and Thomsen3 have shown
that the ultimate load QL is between the values 1.73 and 2.80. The stress field studied by
Kobyashi and Thomsen is such that max (tr (a)l=7 4 ( 3 ) c .The authors' gave a numerical stress
field a,,with Q(a,,)= 2.36 and max Itr (c,,)! ~ 7 J ( 3 ) cMaxltr
. (u*)lwould be probably about
7 4 ( 3 ) c for an optimal admissible u* stress field. In the following we have taken ko = 15c.
The (8) problem has been studied for 5 meshes Mi ( i = 1 , . . . , 5 ) (Figures 6 and 7 ) . The
parameter characteristics of these meshes are shown in Table I and the numerical results are
given in Table 11.
To limit the numerical size of the problem dealing with the mesh Ms, we have studied the
following mechanical problem: compression of half cylindrical sample ( H / 2 ,R ) with a frictional
interface AB and a smooth interface EF* (Figure 8), which is equivalent to the initial problem.
Any optimal admissible velocity field U' for the former leads to an optimal admissible velocity
field u for the latter (Figure 8 ) . So, we have chosen for the initial problem only the velocity
fields verifying the conditions of symmetry defined on Figure 8. Moreover, this symmetry
* Smooth interface: the P.A. stresses are such that: a,,3 0 ;IT[ = 0; the normal Row rule implies: [ U,] = U,, - U,,, 0.
118 S. TURGEMAN A N D J. PASTOR

0 K K

0 K

M3 M4

Figure 6. The meshes M i

C
2

Figure 7. Optimal velocity field for M5

Table I. (mi = 26)

Computer
Mesh NT NDE A(m,n ) I (computing code)
- -. ~

MI 8 4 (40,230) 10.75 IBM 360 (MPS)


M2 16 4 (72,45 0) 23.30 IBM 360 (MPS)
M3 32 8 (136,882) 91.21 IBM 360 (MPS)
M4 64 16 (272,1750) 563-25 IBM 360 (MPS)
M5 192 (96) 8 (400,261 8) 61 IBM 370/168 (MPSX)
NT: number of triangles
NDE:number of discontinuity elements
A ( m , n):matrix of constraints ( m rows, n columns)
1:computingtime in seconds
KINEMATIC APPROACH FOR AXISYMMETRIC PROBLEMS 119

Table 11. (I? = 4, H = 2, c = 2, Uo= 1)

Mesh P(6) P*@) q(v^) jq$ Pb(6) Q Qb

,.
Mi 309.56 296.80 3-01 10.1% 326.90 2.95 3.25
Mz 302.11 288.26 1.14 3.9% 299.66 2.86 2.98
MB 296.15 284.39 1.09 3.8% 295.29 2.82 2.93
296.79 283.55 0.98 3'4% 293.35 2-82 2.91
Ms 291.03 283.88 0.65 2.3% 290.62 2.82 2.89

(though not imposed) is put forward in the fields given by (9')for the cases Msand h14, but
not for these relative to the cases MI and Mz.
Figure 9 shows the variation of ~ ( 6versus
) NT: ~ ( 5 decreases
) linearly from NT= 16.
Though the sets of the kinematically admissible velocity fields given for the different meshes
. are not included; the fact that P(6) (respectively P*(6)) increases slightly from NT=32
(respectively NT = 64) is due to the approximation ' u PA on an average' which involves a

u,
0

Figure 8. Initial problem and equivalent problem

OL.. .
8 16 32 64 192 %T
Figure 9. Variation of ~ ( tversus
) NT
120 S. TURGEMAN AND .
I.PASTOR

I\

Figure 10. Variation of p(t?),P(C)and P,&) versus NT

violation of the external approach of the limit loads by the numerical method. For the MI,
material this method keeps, however, an external approach of the limit loads and PI,(6)
remains strictly decreasing (Figure 10).
The program has not allowed us to improve the upper bound Q = 2-80 given by Kobayashi
and Thomsen. The best result is 0 = 2.82 which is very close to the preceding value. The
mesh being very thin, it seems that this value may be close the the very limit load.
The indirect optimization of P*(6)leads to worse results than direct optimizaion. To balance
this it is necessary to choose the most accurate yield criterion linearization. In this sense,
f i = 26 seems to be a convenient value which does not lead to an oversized matrix A. The
ratio RM(measuring the accuracy of the linearization):

where pd(8) (respectively P: (6)) is the dissipated power in 6 for the ML criterion (respectively
Von Mises criterion) inside the strained zones, is about 6 per cent (for a f i = 11 linearization,
R M is about 23 per cent).
Finally, when restricting 6 to the domains 01 = CI, 0 2 = C2 U C2,. . . ,0
i= CIU . ..U Ci
(Figure ll), we can deduce from the results obtained from MS the loads 0 corresponding to
ratios H/R>0.5 (cf. Table 111).

Figure 11. Domains C,


KINEMATIC APPROACH FOR AXISYMMETRIC PROBLEMS 121

Table I11

H/R -
*(')
P*(6)
Q Qm NT

4 4.8% 1.82 1.79 24


2 4.5% 1.94 1.90 48
$ 4.1% 2-01 2.03 72
1 3.6% 2.21 2.18 96
$ 3.1% 2.35 2.33 120
$ 2.9% 2.50 2-48 144
$ 2.5% 2.66 2.64 168

QKT: results given by Kobayashi and Thomsen3

4.3. Stability of a circular excavation


We consider here a circular excavation of radius R and height H, in a standard material of
unit weight y (Figure 12).
The boundary conditions are the following:
on Ar and BC: a,= 0 ; 7rz= 0 ; 7,e =O;
on AC: a,= 0 ; 7,, = 0 ; T+ = 0 ;
at infinity: ur = uL= U g = O .
The problem involves one loading parameter 0 = yH/c with the associated

It has been studied by Reynaud' who obtained: 2 s QLs 7.2 (QL: ultimate load) for H = R.
In order to determine an approximate value of ko, we used, besides the statical approach
of Reynaud giving 0 = 2, a numerical stress field determined by the authors' for which
0 = 3.06.The upper value for Itr (a)lin the area 2, is 3 . 2 ~for each case. We choose ko = 5c
to estimate the value of A(6).
The computations concerning this problem are performed from meshes F1 and F2 (Figure
13). The meshes are limited to the triangle ADC;an extended meshed zone shows that beyond
the triangle ADC the material has a rigid behaviour. CD is a potential kinematical discontinuity
line.

I R

Figure 12. Circular excavation


122 S. TURGEMAN AND J. PASTOR

Figure 13. Meshes F1 and F2

Table IV presents the characteristic parameters of the meshes, the numerical aspects and
the results (see Figure 14) for this case. The kinematic singularity at the point C (the velocity
at C tends to cause this point to penetrate the medium) does not affect the result and can be
neglected. Indeed, if an excavation of height H + E (Figure 15) is considered, the field 6,

Table IV. (H= R = 1; c = 1; id


= 5; tii = 26)

Mesh F1 F2
NT 49 121
NDE 7 11
A h , n) (213,1349) (509,3305)
t 824 s 300 s
computer IBM 360167 IBM 370/168
(computing code) (m) (MPSX)
46) 27-36 27.26
P*(G) 26-36 26-21
A( d)/P*(6) 1.5% 0.9%
Q 5.27 5.24
Qb 5.34 5-29
KINEMATIC APPROACH FOR AXISYMMETRIC PROBLEMS 123
D

Figure 14. Optimal velocity field for F2

defined in the triangle ACD, no longer reveals any singularity at C and leads to an upper bound

When E tends towards 0, Q ( E )tends towards 0 =P*(;)/&.Finally it should be mentioned


that the kinematic approach of Reynaud' has been appreciably improved.

Figure 15. Kinematic singularity at the point C


124 S. TLJRGEMAN AND I. PASTOR

5. TRESCA MATERIAL CASE

5.1. Linear formulation of the kinematic method


The following deals with the changes to effect when using (9')
for the Tresca material.
For the Tresca material, the P.A. stress domain 9 is defined by the following inequalities:
J[(ur-uz)2+472,] - 2k s 0

with the cohesion k.


Let i be the following linear isometric transformation:

f : S = ( X ,Y , Z , T ) R 4 + a :

9 = = L-'(9), defined by (29):


Y2+T2==2k2
Y 2 +T 2 < U 2 U = 2 4 ( 2 ) k+ Z J ( 2 ) - X CJ 2 0
Y 2 + T 2 SV 2 V=2J(2)k-ZJ(2)+X VaO
is included inside the polyhedral domain 9 ; defined by (30)
G1,j(S)= Y cos ai + T sin aj - J(2)k s 0
G2,(S)= Y c o s a j + T s i n r r j - 2 ~ ( 2 ) k - J ( 2 ) Z + X ~ O
G 3 , j ( S ) =Y cos ai+T sin a i - 2 J ( 2 ) k + J ( 2 ) Z - X s O
rrj =2nj/fi j = 1 , . . .,fi
9 ; is the PA stress domain for the linearized Tresca criterion (LT criterion).
For a standard material obeying the L.T. criterion, a PA u is given by:
KINEMATIC APPROACH FOR AXISYMMETRIC PROBLEMS 125

(31 cont)

where the ALi ( I = 1,3, j = 1, Ijl) are positive real values as:

(31) is equivalent to:

where the cp:, cp:, cp:z are linear expressions of AIJ.


The dissipated power in tl is:

The constraints between ALi and the nodal variables of a triangle Ti: Urk, uzk, are similar to
the constraints (17): from (33), they have the form:

cpiz(A f.j> = ( Y ( u r k ) + B ( U z k ) ) / 2
(The Af,i relative to Ti are denoted hii).
The condition (34) (identical to (11)) is replaced by the condition (16): u P.A. on an average.
The dissipated power on Ti, PLT ( u ) is given by:

PLT(u) = ./(2)k2nr&, min (11 A ij) (37)


i l
The power dissipated in the discontinuities and the constraints imposed on PA velocity
fields, in these discontinuities, are identical for the Von Mises and Tresca materials (k takes
the place of c ) .
Finally, as in section 4.1, we can define an incertitude due to the approximation u PA on
an average. In the same way we get:

k& = max Itr (a*)l optimal admissible stress field)


(a*:
Z,

5.2. Tests
5.2.1. Ultimate compressive strength on a cylindrical sample. We consider here the mechanical
problelm, Section 4.2, with a k Tresca sample. When using the mesh M5we get, thanks to
the Tresca programm (cf. Table V):
6:the optimal field;
@(iJ): the dissipated power for the LT criterion in iJ;
P(t7):the dissipated power for the Tresca criterion in I?;
126 S. TURGEMAN AND 1. PASTOR

Table V

Tresca model Von Mises model


R = 4 , H = 2 , Uo= 1, k = 2 , tii = 8 R = 4 , H =2, Uo= 1, ~ = 2tii=26
,
A(400,2426) A(400,26 18)
t,= 65 s (IBM 3701168 code MPSX)f-= 61 s
P(0')=318.86 P(B)= 297.03
P'(0')=313*237 P*(B)= 283.88
&=Pr(0')/(id= 3*11* QM = P*(C)/qd = 2.82
pM-T
PT+M(d) =283*575 ( u ) = 313.606
A

~ ( 0 '=
) 0.63 ~ ( 6=) 0.65
--
A(v') - 2.0y0 --
'(6) - 2.3%
P'(0') P*(B)

* The upper bound given by Reynaud' is & = 3.24.

PT+M(v'): the dissipated power for the Von Mises criterion in v' (a tr P.A. field for Von
Mises is P.A. for Tresca, and reciprocally).
In Table V, we recall the results given by the 'Von Mises model' and we give the value of
P''T(6), dissipated power in 6 for the Tresca criterion.
The set of Tresca plastically admissible stress tensors includes the set of Von Mises plastically
admissible stress tensors when c = k ;so we have QT > QM.
The ratio R M is strongly greater than RT though the Tresca linearization leads to 24 A scalar
values instead of 26 for the Von Mises linearization. The preceding shows clearly that the
Tresca linearization is more accurate than the Von Mises one, and explains why PT'M(C)is
lower than P*(i?).
We can deduce from the performed test, the results identical to those of Table I11 in the
case of Tresca material (cf. Table VI).

5.2.2. Stability of a circular excavation. We treat now the problem described in Section 4.3
with a material obeying the Tresca yield criterion (cohesion k). The parameters of the numerical

Table VI

HIR QT NT
4 2.09 24
2 2.21 48
f- 2.35 72
1 2.49 96
$ 2.63 120
$ 2.79 144
$ 2-95 168
KINEMATIC APPROACH FOR AXISYMMETRIC PROBLEMS 127

problem and the results obtained when using the F2 mesh are the following:
A ( m , n) = A(509,3063)(6 = 8)
t = 179 s (on IBM 370/168, code MPSX)
P(0)= 28-35 (k = 1, ( i d = 5)
P'(r7)= 27.76
QT=5.55
A(O)/P'(O) = 0.85%
As previously observed in Section 5.2.1., PT"M(r7) and PM'T(d)are nearly identical respec-
tively to P*(;)and P'(r7). The functionals 'dissipated power' relative to each of the two
linearized criteria have a similar behaviour: the velocity field which minimizes the first
functional nearly gives minimal the latter and reciprocally.

6. CONCLUDING REMARKS
The numerical kinematic method presented in this paper does not lead to a rigorous external
approach of F(K),related to the fact that the strain rate fields are P.A. only on an average
in the strained zones.
The work presented here provide an evaluation of this approximation. In Section 4 and
Section 5.2 it is shown that results of acceptable accuracy can be produced.
In spite of the average condition for u P.A., the number of equations is high compared
with the number of nodal variables: the problem involves a reduced degree of freedom. So,
from a mathematical point of view, it is necessary to include discontinuities in the meshes, as
suggested by experimental data. Extension of this degree of freedom can be carried out by
changing the relationship (16) into the condition: tr u (Gi)+ tr u (Gj)= 0, where Giand Gj denote
the centres of mass of two adjacent triangles. Calculations performed with the latter condition
lead to a value of q ( v ) 10 times greater than the value obtained with the condition (16). For
instance, the calculation of Section 4.2 gives (case m): P(6)= 290.91; P*(6) = 277.02; ~ ( 6=)
9.15, which lead to an unacceptable relative error up to 33 per cent concerning the value of
P*(6) with ko= 15c.
Finally the comparison between the Von Mises and Tresca material behaviours, which are
identical in the plane strain case is studied here in the axisymmetrical case. The linearized
dissipated powers for both materials are optimal for nearly similar kinematical fields. If this
result may be extended in the general case, after other controls, it will be sufficient to use either
linearized plasticity criterion to get results for the two types of material. To that extent Tresca
material seems more interesting because it allows a more easily refinement of the linearisation.

NOTATION
Q = ( Q l , . .. , Q,) = n parameters loading process
4 = (dl, . . . ,(in)=strain rate parameters
K = convex of the admissible loadings
F ( K )= boundary of K
P.A. = plastically admissible
K.A. = kinematically admissible
128 S. TURGEMAN AND J. PASTOR

S.A. = statistically admissible


u = stress field
2) = strain rate field
u = velocity field
a,, ug, a,,T,,, 7+7ez = stress tensor components in r, 0, z cylindrical coordinate system
t),, Q,, u,, urzrud, uez = strain rate tensor components in r, 0, z cylindrical coordinate
system
un ue,uz = velocity vector component in r, 8, z cylindrical coordinate system

For the Von Mises material


L.M. criterion = linearized Von Mises criterion
c = cohseion
rii = parameter of linearization

P(u)= dissipated power for the linearized criterion


P*(u)=dissipated power
8 = optimal velocity field

For the Tresca material


L.T. criterion = linearized Tresca criterion
k = cohesion
& v ) = dissipated power for the linearized criteiron
P f v )= dissipated power
d = optimal velocity field
rii = parameter of linearization

REFERENCES
1. M. Fremond and J. Salenqm, Analyse limite par une mtthode dtltments finis, Congrks Franqais de Micanique
de Poitiers (1973).
2 . M. Fremond, A. Pecker and J. Salencpn, Mtthode variationnelle pour le mattriau rigide-plastique, Symposium
Franco-Pulonaisde Rhiologie er Micanique des Sols, Nice (1974).
3. S . Kobayashi and E. G. Thomsen. Upper and lower bound solutions to axisymmetric compression and extrusion
problems, Inr. J. Mech. Sci., 7 , 2 (1965).
4. B. Mercier, Une mtthode pour rtsoudre le problbme des charges limites, J. Micanique Appliqub, 16, No. 3
(1977).
5. J. Salenwn, Applications of rhe Theory of Plasticify in Soil Mechanics, Wiley, 1077.
6. J. Pastor and S. Turgeman, Mixen oeuvre num6rique des mtthodes de Ianalyse limite pour les mattriaux de
Von Mises e t de Coulomb standards en dtformation plane, Mech. Res. Comm., 3,469-474 (1976).
7. J. Pastor and S. Turgeman, Limit analysis in axisymmetrical problems: numerical determination of complete
statical solutions, to appear.
8. G. Reynaud. Approche variationnelle de quelques problbmes axisymttriques en MCcanique des solides. 7Wse
de Docrorat de sficialiti, Grenoble University, IMG (1979).
9. J. SalenGon, Calcul ii la rupture et analyse limite, Cours d 1E.N.P.C. (1978).
10. S. Turgeman, Chargernents limites de fondations sollicittes ii Iarrachement: approches statique et cintmatique,
J. Micanique appliquie, 3. No. 1 (1979).

You might also like