finite element analysis of excavation
TRANSCRIPT
Computers and Geotechnics 1 (1985) 207-220
F I N I ~ K I ~ N T ANALYSIS OF EXCAVATION
P.T. Brown and J.R. Booker School of Civil and Mining Engineering
University of Sydney Sydney, N.S.W., 2006
Australia
ABSTRACT
The finite element method has often been used to simulate excavation. When the soll is linearly elastic, the results of excavation should be independent of the number of stages in the excavation process, and lack of such independence indicates an incorrect procedure. The simple direct method described in this paper provides the required independence In the case of linearly elastic materials, and hence can be used for multl-stage excavation in non-linear problems without excessive errors. However methods whose errors increase with the number of stages of excavation are quite unsuitable for non-llnear problems. Alternative methods of analysis, errors arising from the inability of the elements to model adequately the stress gradients near the toe of the excavation and excavation adjacent to a diaphragm wall are discussed.
INTROML~FION
The finite element method provides geotechnlcal engineers with a
potentially powerful tool for simulation of the excavation process.
However, care must be taken in the use of this tool if significant errors
are to be avoided.
For example Christian and Wong in reference i reported some serious
errors discovered while developing a program to simulate construction of
braced excavations. They concluded that excavations should be simulated in
as few stages as possible, because trouble arises when points of stress
concentration are excavated.
While it may be satisfactory to simulate an excavation in linearly
207 Computers and Geotechnics 0266-352X/85/$03.30 © Elsevier Applied Science Publishers Ltd, England, 1985. Printed in Great Britain
208
elastic material by excavation in one stage, this form of simulation is
unlikely to be satisfactory when the material behaviour is non-linear.
Hence the authors of this paper sought a method which did not introduce the
errors referred to above. This method could be checked for the linearly
elastic case as the results should be independent of the number of stages
shown in reference [2]. Such a method is presented here, and as is
intended as the basis for simulation of excavation in non-linear
materials.
EXCAVATION SIMULATION"
The basic aspects of simulation of excavation by the finite element
method are summarised as follows. Fig.l(a) shows a body of soil from which
the shaded portion A is to be excavated, leaving the unshaded portion B.
The behaviour of B will be identical if material A is removed and replaced
by the tractions (T) which were previously internal stresses in the soil
mass, as indicated in Fig.l(b). Then the behaviour of B due to excavation
will be the behaviour of B when the tractions T are removed, for example by
applying equal and opposite tractions, as shown in Fig.l(c).
Simulation of a stage of excavation thus involves determination of
the tractions T at the new portion of the soil boundary, determination of
the stiffness of the soil mass B, and application of tractions -~ to the
new portion of the boundary.
Finite element implementation of this process involves determination
(a) (b) (¢)
FIGURE i. Simulation of Excavation
209
of the noda l f o r c e s which a r e e q u i v a l e n t to the t r a c t i o n s shown i n
Fig. l (c).
This may be carried out by a variety of methods. Christian and Wong
considered methods involving direct determination of tractions and equlva-
lent nodal forces from known values of stress. These gave results which
were dependent on the number of stages of excavation, except when a high
order polynomial was used for stress extrapolatlon. The implementation of
such methods appears laborious, and as Christian and Nong say, the method
may not be applicable to all types of excavation problem.
Clough and Mana in reference [3] report that blana in reference [4]
proposed to determine the excavation boundary forces as
M f = ~ f sTo dV ~ m=l v
in which M is the number of elements which have a common boundary with
unexcavated elements, B is the displacement strain matrix, o is the stress
vector, and f is the vector of nodal forces. However, Desal and Sargand
in reference [5] report that this method is unable to produce correct
results for a simple problem involving excavation off the top of a single
column of elements. Desal and Sargand propose a method using a hybrid
stress model, which is also based on integration of stresses, but includes
integration of surface tractions on the boundary of the domain.
An alternative approach, which is the basis of this paper, is to start
from the virtual work equation including the terms due to stresses, body
forces and external tractions. Then as is shown in the next section of
this paper, the nodal forces can be found by appropriate numerical integra-
tion of stresses, body forces and external tractions throughout the soil
mass, in such a way that total equillbrit~n is approximately maintained at
each stage of excavation. While methods based on extrapolation of known
stresses are conceptually simpler, the recommended method is easier to
210
implement ~Ithout loss of accuracy. Other integration methods, which
ignore part or all of the contribution of body forces to the total strain
energy, will in general give incorrect results, although they may provide
adequate solutions to some problems.
FI]IITE ELEMENT IMPI..EMlmT&TXON
Let us consider a multi-stage excavation for which the inital stresses
are no' and the initial strains are ~o~ and are zero as shown in Fig.2.
The soil mass is subject to body forces ~ and surface tractions t outside
the excavation, and after the i-th stage of excavation the stresses and
strains are ~i and ~i respectively. V i is the volume of the soll mass
mass after the i-th stage of excavation has taken place, and it is required
to determine the nodal forces which must be applied to this volume in order
to simulate satisfactorily the process of the i-th stage of excavation.
t t i i l i i / I I
Vo---
~ : .o - o
.~o: O
Initk~l s t o r e J
V~
g=.~l
-.~o = D .~1
End of f i r s t s tage
t t i 1 t
_~ = .~2
O'2- O1 = D (e2-el)
End of second s tage
FIGURE 2. Definition of stresses and strains at each stage.
211
The procedure adopted for determination of these nodal forces is to
impose an arbitrary (virtual) nodal displacement at the end of the i-th
stage of excavation. Then by equating the internal and external work
performed during that displacement, the nodal forces required to simulate
the i-th stage of excavation can be determined, based on total not
incremental equilibrium.
Let the virtual displacement be 6u and the corresponding strains and ~
nodal displacements be 6e and ~a. Equating the internal work done by the
stresses and the external work done by the body forces ~ and the tractions
f(6~T~i)dV = f (~uT!)dV + f (~T!)dS V i V i S i
(1)
where S i is the surface of the soll mass at the end of the i-th stage of
excavation.
Now writing
A~i ffi ~i - ~i-I and taking ~o = 0
A~i ffi ~i - Zi-1
Ao i~ = DAe i~ where D is the incremental stress-strain matrix
equation (i) can be re-written as
f (~cT(Aoi + Zi_l))dV = f (~uT~) dv + f(6~T~)dS Vi ~ ~ Vi ~ ~ Si
f(6~TDA~i)dV = - f (6Toi_ l )dV + f(6uT~)dV + f(eTt)dS V i V i V i ~ ~ Si
(2)
f(~aTBTDBAai)dV ffi - f(~aTBToi_l)dV + f ( 6a~T ~) dV + f(~aTNT~)d$ V i V i V i S i
where N is the mat r ix of shape f u n c t i o n s .
212
Since the above relationship holds for arbitrary displacements
(~(BTDB)dV)Aa = -f(BToI_I)dV+f(NTy)dV + f(NTt)dS
Vi ~ Vi - _ V I ~ Si ~
or in finite element form
K'A~i = [ i , i - 1 + ~i + ~i (3)
where ~i'i-l' ~i' ~i are the vectors of nodal forces arising from integra-
tion of ~i-l' ~ and ~ respectively and the first subscript indicates the
the body over which the integration takes place, and the second subscript
(if any) indicates the stress state which is integrated.
The sum of the vectors f, g and h ~ represents the nodal forces which
need to be applied to retain total equilibrium in carrying out the i-th
stage of excavation.
Restricting attention now to the linearly elastic case, it is of
interest to rewrite equation (2) in the form
f(6~ TD ~i)dV ffi -f6eT(Oi_l - D~i_l)dV + f(6uT~f)dV~ ~ + f(6uTt)dS~ ~
V i V i V i S i
(4)
From Fig 2 it will be seen that
~i-I - D ~i-I = ~o° (5)
and making this substitution in equation (4) and converting to finite
element form
K'~i ffi ~i,O + ~i + ~i (6)
Equation (6) corresponds to a single stage excavation, and is
precisely equivalent to repeated application of equation (3) for the
linearly elastic case, provided that the sample points used to evaluate the
stiffness matrix are also used to evaluate the nodal forces corresponding
to each 'initial stress'. Thus while there will be errors in the results
213
obtained due to inadequate representation of the stress distribution by the
finite elements, no additional errors will be caused by increasing the
number of excavation stages, if this method is used. Hence this method
achieves the theoretically correct equality between single-stage and
multi-stage excavation.
DISCUSSION OF RXSULI'S
Finite element analyses were carried out using 9 Gauss points per
element and the meshes shown in Fig.3, which consist of 8-noded
smoot h ~
(a)
smooth~
(b)
smooth~__
(c)
P,| Im
• 0 1"0 1-0 1"0 1'0 1"0 1'0 1"0
1"0
1'0
1.0
1"0 fixed
0.1k~7 1'0 1.0 1-0 1.0 1-0
o.1 0.2
not to scale
Y/) IIII I I I I
" ' I ; ' , ;~1111 ' , I I I I I I I I
1.0
0.9 0.1 ~--------0.1 " ~ ' 0 . 2
1.0 " ' f i x e d
0"72
~ 0 < ) 6
. . . . . ,,,, ~ 0 " 0 6
III111111 1.0 " 0 . 7 2 0 1.0 1.0 1-0 1.0 " ' f i x e d
b o6cb(b(b66
FIGURE 3. Finite element meshes used.
214
isoparametric rectangles, and where the material was homogeneous isotropic
and linearly elastic. The excavated elements have been cross-hatched, and
the excavation was carried out in one-stage, or in several stages each
consisting of one horizontal row of elements. In each case the nodal
displacements obtained at the completion of excavation were identical,
whether the excavation was carried out in one or several stages. For all
the analyses discussed in this paper Young's modulus (E) was I0,000 kPa,
Poisson's ratio (v) was 0.47, K o = 0.5 and the unit weight of the soil
was 20 kN/m 3 .
Analyses using the mesh shown in Fig. 3(a), were also carried out
using 4 Gauss points per element, in which the nodal forces were determined
by linear extrapolation of stresses from the Gauss points in the elements
next to be excavated. This was followed by fitting a straight line to the
resulting values of stress on the excavation boundary for each element, and
/l/'/~'2' Proposed Method , / / ~ Stress Extrapolation:
/:~ / 2 Stage. 0.Sm Elements o o o J J I 2 Stage, ~Om Elements . . . . . " k ~ 4 Stage, C)-Srn Elements . . . . . . .
1 " - ~ I I 4 3 2 1 0
Horizontal Displacement (mm)
FIGURE 4. Displacements of side of cut using stress extrapolation.
215
calculation of the equivalent nodal forces. Two similar analyses were
carried out with the two upper rows of elements in Fig. 3(a) subdivided
into elements each of height 0.5 m, in order that the excavation could be
carried out in 4 stages and also in 2 stages of two rows of elements. The
horizontal displacements of the side of the cut resulting from use of the
recommended method, and the analyses based on stress extrapolation are
shown in Fig. 4. Excavation in one stage produced the same results as the
recommended method, but progressively greater errors are seen to occur,
using the stress extrapolation method, when the excavation is carried out
in 2 stages and 4 stages. Similar increases in error with an increasing
number of excavation stages were reported by Christian and Wong when
evaluating nodal forces only from elements near the excavation (not using a
high order polynomial for extrapolation). However it is of interest to
note that at most depths there is a smaller error for 2 stage excavation
with the finer mesh than with the coarser mast.
Nodal forces were found to arise only on the excavated boundary of the
current stage when the recommended method was used. When the contribution
to the nodal forces arising from the body forces is ignored, vertical
forces arise at every node in the soil mass, and produce grossly erroneous
results. However if one also ignores the nodal forces except at the nodes
on the current excavated boundary, a reasonably satisfactory solution is
obtained. Table i shows the vertical nodal forces on the boundary to be
excavated in the first stage of excavation using the mesh shown in
Fig. 3(a), and 9 Gauss points per element. The contributions arising from
integration of stresses and body forces are shown separately, as well as
the total nodal forces for each of the relevant nodes, whose numbers and
locations are shown in Fig. 5.
216
TABLE i Vertical Nodal Forces (kN)
Node fBTo dV fNT~ dV Total
7 13 21 27 35 36 37
1.67 20.00 3.33
20.00 -1.67 6.67
-1.67
1.67 -6 • 67 3.33
-6.67 5.00
-6.67 1.67
3.33 13.33
6.67 13.33
3.33 0 0
Total 48.33 -8.33 40.00
0
.•0.5
0
o
--Node Numbers ( ~ o 1 2 3 4 5 Cram) / ~
L ® ® ®
1.5 I (>5 0
Distance f rom Side of Cut (m)
FIGURE 5. Displacements at end of excavation.
It can be seen that ignoring the body force contributions, leads to
significant error in the vertical nodal forces. Similar errors occur in
the second stage of excavation. Refinement of the mesh by subdividing the
upper two rows of elements into elements of height 0.5 m, had no
significant effect on the final displacements. Despite these errors in
217
vertical nodal forces, the final displacements of the boundary of the
excavation resulting from use of the recommended method, and stress
integration ignoring body forces and non-boundary forces, produce similar
displacements as shown in Fig. 5. The recommended method gives the correct
solution for the selected mesh, since it is precisely the same as the
solution for one-stage excavation using this mesh. However this stress
integration method provides a solution which is reasonably similar except
for the vertical displacements near the top of the cut.
mesh in f c / / a
i i
4 3 2
Horizontal deflection (mm)
I I O
FIGURE 6. Horizontal deflections of side of excavation when using 9 Gauss points.
A very coarse mesh such as that shown in Fig.3(a) cannot be expected
to model the stress concentrations at the bottom corner of the excavation.
Better representation of the stress distribution in this region, by the use
of a finer mesh near the corner, should produce more accurate values of
displacements. This aspect was investigated by means of additional finite
element analyses using the meshes shown in Fig.3(b) and (c). The resulting
218
horizontal nodal displacements along the vertical excavated face for the
case of 9 Gauss points are shown in Fig.6. The first mesh refinement leads
to an increase in deflection of the top corner of the excavation of 15%,
however the second mesh refinement only leads to an additional increase in
deflection of 3%.
It is also of interest to examine the effects of changes in element
size when the excavation is carried out adjacent to a diaphragm wall. For
this purpose finite element analyses were carried out on the meshes shown
in Fig.7 which include a concrete diaphragm wall. The wall deflections
above the base of the excavation are sbown in Fig.8.
/ srnooth
~, i 1 '0
L ~ f i x e d 1.0 1.0 0.5 1.0 1.0 1.0 1.0 1.0 1.0
f smooth.
0 ~ 0 1-0 1.0 1.0 1.0 1-0
• 1 0.5
1.O
1.0 ....~ 0-1 ~ 0 . 2 0.7
1.0 " ' f i x e d
FIGURE 7. Finite Element Meshes used for diaphragm wall.
The effect of refining the mesh is smaller than when no wall was
present, as the increase in deflection at the top corner of the excavation
is only 10%.
219
Finer mesh
/ 4 Gauss points / / 9 Gcuss points
I I I I 4 3 2 1 0
Well deflection (ram)
FIGURE 8. Deflections of diaphragm wall.
CONCLUSIONS
A theoretically sound basis has been established for a simple correct
method of finite element simulation of excavation. This produces results
which are independent of the number of stages in which the excavation is
carried out for the case of a linearly elastic medium. The implementation
of this method appears to be more economical to program than the high order
polynomial fitting proposed in reference [I], aud is more accurate than
other suggested methods. The method described is therefore recommended as
the basis for simulation of excavation in non-llnear materials.
It has been demonstrated that for a linearly elastic material, very
little refinement of the mesh is required in order to obtain horizontal
deflections of engineering accuracy.
220
ACKIIOIIL~IqTS
The work described in this paper was carried out with the aid of a
grant from the Australian Research Grants Committee.
REIzgRENCKS
i.
2.
3.
4.
5.
Christian, J.T., and Wong, I.H., Errors in simulating excavation in elastic media by finite elements. Soils Fdns. (Japan). 13 (1973) i-i0.
Ishihara, K., Relations between process of cutting and uniqueness of solutions. Soils Fdns. (Japan). IO (1970) 50-65.
Clough, G.W., and Mana, A.I., Lessons learned in finite element analysis of temporary excavations. Proc. 2nd Int. Conf. Num. Meth. Geomech., A.S.C.E. I (1976) 496-510.
Mana, A.I., Finite element analysis of deep excavation behaviour. Ph.D. Thesisp Stanford University, Stanford, Calif. (1976).
Desai, C.S., and Sargand, S., Hybrid FE procedure for soil-structure interaction. J. Geotech. Ensg. , A.S.C.E. iio (1984) 473-486.