Вы находитесь на странице: 1из 13

Computers and Geotechnics 41 (2012) 2335

Contents lists available at SciVerse ScienceDirect

Computers and Geotechnics


journal homepage: www.elsevier.com/locate/compgeo

An upper bound-based solution for the shape factors of bearing capacity


of footings under drained conditions using a parallelized mixed f.e. formulation
with quadratic velocity elds
A.N. Anto a,, M. Vicente da Silva a, N. Guerra a, R. Delgado b
a
UNIC, Department of Civil Engineering, Faculty of Sciences and Technology, Universidade Nova de Lisboa, Quinta da Torre, 2829-516 Monte da Caparica, Portugal
b
Technical Department, Sociedade de Construes Soares da Costa, Grupo Soares da Costa, SGPS, SA, Rua de Santos Pousada, 220, EC Municpio, Apartado 4862,
4000-101 Porto, Portugal

a r t i c l e i n f o a b s t r a c t

Article history: A strict upper-bound limit analysis nite element formulation is used to estimate shape factors sc and sq
Received 24 June 2011 for determining the bearing capacity of shallow foundations using the classic bearing capacity formula.
Received in revised form 28 October 2011 The nite element formulation uses a quadratic approximation for the velocity eld, an extension of a
Accepted 2 November 2011
previously published Augmented Lagrangian formulation with a linear velocity eld, and was imple-
Available online 17 December 2011
mented for a parallel processing environment. Results from determining the limit loads under three-
dimensional conditions are presented and compared with previously published data. The results obtained
Keywords:
allow a strict upper-bound determination of the shape factors. Furthermore, a practical proposal for these
Footings
Shape factors
factors is made and compared with other proposals made by other authors.
Parallel processing 2011 Published by Elsevier Ltd.
Quadratic elements

1. Introduction Shape factors sc and sq are often used to correct (2) in order to
take into account the three-dimensional effect on the bearing
The bearing capacity of a strip footing with width B acted upon capacity, due to the shape of the footing. For rectangular footing
by a vertical centered load can be expressed by Terzaghi [1] with width B and length L P B this equation becomes

qB1
u 0:5cBN c c0 Nc qN q 1
qBL
u qBL BL
u;c qu;q 0:5cBN c sc qN q sq 3
0
where c is the soil unit weight below the footing base level, c is the
effective cohesion, q is the surcharge at the level of the footing base, Several classical proposals [25] have been made for the values of
and Nc, Nc and Nq are bearing capacity factors which are functions of these shape factors. More recently, this problem has been addressed
the soil friction angle /0 . by other authors [69]. The present paper intends to contribute to
For a cohesionless material, (1) becomes the determination of shape factors and will use these recent works
for comparison. A strict upper bound limit analysis nite element
qB1 qB1 B1 formulation, employing quadratic approximations for the velocity
u u;c qu;q 0:5cBN c qN q 2
eld, will be used. All results obtained in the present paper therefore
Eq. (2)as well as (1)is an approximation. In fact, the two terms assume an associated ow rule.
of (2) are solutions to two different problems: the rst term, In order to achieve estimates for the shape factors with a signif-
qB1
u;c 0:5cBN c , assumes a null surcharge q; the second one, icant level of accuracy, a parallel computing approach was used.
qB1
u;q qN q , is obtained by considering the soil unit weight below Such a framework allows of signicantly increasing the number
the level of the base of the footing to be equal to zero. The two effects of degrees of freedom involved in the analysis and, consequently,
are superposed but this is not theoretically correct, although it is improving the quality of the results.
traditionally accepted and it will also be accepted in the present All the computations reported in this paper were performed on
paper. a small scale cluster. This cluster comprises eight nodes, each
equipped with two Intel Xeon 2.00 GHz quad-core processors
Corresponding author. Tel.: +351 212948580; fax: +351 212948398. and with 16 GB of RAM memory. The cluster nodes are intercon-
E-mail address: amna@fct.unl.pt (A.N. Anto). nected with a dedicated 2  Gigabit Ethernet network.

0266-352X/$ - see front matter 2011 Published by Elsevier Ltd.


doi:10.1016/j.compgeo.2011.11.003
24 A.N. Anto et al. / Computers and Geotechnics 41 (2012) 2335

2. Formulation In (6)(7), the domain X is delimited by the boundary C, com-


posed of its two complementary parts (C = Cu [ Cr ^ Cu \ Cr =
2.1. The limit analysis problem ): the kinematic boundary, Cu, and the static boundary, Cr,
where displacements are xed and the external surface forces are
Limit analysis (LA) theorems have been employed extensively prescribed, respectively.
in the development of numerical models used in the direct estima-
tion of the collapse loads of structures. In order to be in the domain 2.2. Mixed nite element model
of application of LA theorems, the structure must:
As far as we know, all formulations found in the literature that
 exhibit a perfectly plastic behavior, expressed as usual by estab- are able to produce strict upper bounds, in the sense that the col-
lishing an immutable yield surface, for the whole body domain lapse mechanisms obtained must be fully compatible (namely,
X, constraint (6e) must be veried locally), use at most quadratic
f r 0 in X 4 velocity elds based on triangular or tetrahedral element meshes,
for 2D and 3D cases, respectively. Such is the case for the formula-
Associated with any state of stress, r, inside the yield surface the tions reported in the works of [1316], just to cite a few of the most
material response is elastic. When the state of stress lies on the sur- recent and relevant papers on this eld. The limitation of using
face, the material enters a plastic state and plastic ow can occur. velocity elds with polynomials of no higher than the second de-
The remaining states of stress, the ones located outside of the sur- gree is related to the mathematical complexity of fullling locally
face, are considered to be inadmissible. the associated ow rule for any other kind of velocity elds, which
 exhibit an associated plastic ow rule. The plastic strain rate, e_ , is a necessary condition to ensure the strict validity of the upper
direction must be oriented according to the outward normal of bound theorem.
the yield surface. In this work, this condition is expressed math- In this section, we present the extension of the Augmented
ematically by the constraint, Lagrangian formulation for linear velocity eld elements proposed
in [17,18] for the case of quadratic elements, preserving the capac-
e_ 2 Cc 5 ity of generating strict upper solutions of the original model. More-
over, particular attention was paid to not jeopardizing the
where Cc denes the space of all strain states normal to the yield
numerical effectiveness of the non-linear minimization procedure
surface (4). Mention must be made of the fact that, from the theo-
performed element-by-element, as described in [17]. In a concise
retical point of view, the elastic strains produce no effects on the
way, the effectiveness of this minimization is due to the fact that
load bearing capacity of the current kind of structures [10,11]. For
it is performed directly, without any iterations. This aspect is most
this reason, elastic strains are disregarded in this work and the
relevant for the overall performance of the Uzawa based iterative
material is treated as if it was rigid-plastic.
solution method, because, although the number of local element-
~ kt ~t, based minimization is increased, it is still possible to perform it
The structure is subject to a given set of loads, kb b;
~ and t or ~t denote, respectively, a body and surface directly.
where b or b
As a starting point, the optimization problem (6) is discretized
forces distributions with a xed amplitude, and where k is the
using the mixed nite elements outlined in Table 1. These ele-
so-called load multiplier, a positive scalar. Body and surface forces
~ and ~t are, therefore, always xed, whereas kb and kt are variable. ments are regarded as mixed since, in the nite element, the veloc-
b
ity and the strain approximations are assumed to be independent
It is a well established fact that the LA theorems can lead to
from each other. Thus, the two eld approximations are:
mathematical optimization problems in order to obtain either a
lower bound (using the static theorem) or an upper bound (using
u_ i Q i d in Xi 8
the kinematic theorem) of the load multiplier associated with the
i i
structures maximum load bearing capacity. The duality between _ i
e L e in Xi 9
these two approaches has also been established (see, e.g., [12]).
To proceed, the following mathematical programming, based on As regards the (i)th nite element, matrices L and Q collect,
the kinematic theorem, is cast: respectively, the conventional linear and quadratic nodal approxi-
Z mation functions (expressed as usual in the element local coordi-
minimize k e u
De_ dX  P _ 6a nate system) [19], and the vectors d and e are their associated
X weights, given by the nodal velocity and plastic strain rate values.
subject to u_ 0 in Cu 6b In the case of 3D problems this is

e_ Bu_ 6c u_ i fv x ; v y ; v z gTi I33 Q 1 ; . . . ; I33 Q 10 i d 10a


_ 1
Pk u 6d _ i Ti
e fe_ x ; e_ y ; e_ z ; 2e_ xy ; 2e_ yz ; 2e_ xz g i i
I66 L1 ; . . . ; I66 L4  e 10b
e_ 2 Cc 6e
where I[nn] denotes the identity matrix of n  n dimension.
where u_ denotes the velocity and B the standard differential com- It should be pointed out that the velocity eld approximation
patibility operator. D function returns the plastic energy dissipation (8) is a global one, since all elements sharing a common node, n,
rate per unit volume, while P e and Pk represent the work rate of the also share the same associated nodal velocity, dn. In contrast, the
external loads, expressed by the following linear functionals: strain rate eld approximation (9) is a local one: in fact, each ele-
ment has its own independent strain weight vector, e.
Z Z
T The nodal linear approximation functions are also employed as
Pk b u_ dX t T u_ dC; 7a
X Cr
shape functions, thus preventing the distortion of the approxima-
Z Z tion functions when mapping the local coordinate system into
e
P ~T u_ dX
b ~tT u_ dC; 7b the global Cartesian referential system. For this reason, these qua-
X Cr dratic elements can be considered as subparametric [20].
A.N. Anto et al. / Computers and Geotechnics 41 (2012) 2335 25

Table 1
Finite element approximations.

Dimension Shape functions u_ approximation e_ approximation


2D

3 nodes 6 nodes 3 nodes


3D

4 nodes 10 nodes 4 nodes

To proceed, following the previously adopted strategy [17], the with


Z Z
Lagrange multiplier method is applied to the optimization problem i
(6) in order to introduce the constraints (6c) and (6d) into the
pi LTi BQ i dX d  LTi Li dX ei 16
Xi Xi
objective function. This leads to a minimization problem expressed
as: This approach consists of the so-called Augmented Lagrangian
Z method. The p term, also present in the functional L0 denition
Min _ e_ ; l; lk
Lu; e u
De_ dX  P _ lk 1  Pk (13), is responsible for establishing the relationship between the
e_ 2Cc X global and the local strain elds. Resorting to Gaussian quadrature
Z
formulae, its volume integrals can be expressed in an exact manner
l : Bu_  e_ dX 11
X by sums:
!
Di X
nN
where lk and l vector collect the Lagrange multipliers. pi LTi xn BQ i xn d
i

Using the eld aproximations (8) and (9), it is possible to obtain nN n1


the following discrete Lagrangian form of this problem: !
Di X
nN
Ti i
 L xn L xn ei 17
min max Ld; e; lk ; l L0 12a nN n1
d;e lk ;l
subject to ei
n 2 Cc ; n 1; . . . ; nN 12b where D represents the element volume and xn denotes the natural
coordinates of the nth Gaussian integration point.
where the Lagrangian functional yields
nE Z Remark 1. An attentive inspection of (17) leads to the conclusion
X
L0 DLi ei dX that the compatibility between the global and the local strain elds
i1 Xi is imposed uniquely at the nN Gaussian points, for each nite
Z Z ! element. However, since the plastic strain eld approximation is a
X
nE
 ~T Q i dX
b ~t T Q i dC di linear one, the previous condition is sufcient to guarantee (6c)
Xi Ci
i1 r throughout the element domain.
" Z Z ! #
X
nE
T i As presented up to this point, this formulation can be regarded
lk 1  b Q i dX tT Q i dC d
Xi Ci as a straightforward extension of the model reported in [17,18]
i1 r
Z Z  for quadratic elements. However, when attempting to implement
X
nE
lTi LTi BQ i dX d 
i
LTi Li dX ei 13 the same solution procedure it was observed that this was no longer
i1 Xi Xi possible [21], due to a coupling of the strain variables, e, associated
with the distinct nodes of the element. The source of this coupling
in which nN and nE denote, respectively, the number of nodes per
was tracked down and identied as a consequence of the second
element and the mesh number of elements; lk and l are the
term of (17).
Lagrange multipliers variables. It can be shown as in [21] that the
Fortunately, the Remark 1 led to the solution of this problem. In
Lagrange-multiplier eld,
fact, it can be easily concluded that, without violating the compat-
Li li ri ; 14 ibility, the global and local strain bond can be established using
any set of nN non-collinear points belonging to the nite element
represents the dual of the strain eld (9), or, in other words and domain. Thus, the Gaussian points were replaced, in (17), by the
from a physical point of view, the stress eld. points located at the vertices of the element. By denition, the va-
As discussed in [14], if, for a nite element with a linear plas- lue of the approximation function, Ln, associated with vertex n
tic strain eld approximation, the plastic strain state at each ver- equals 1 at this vertex and is null at all the other vertices. Hence,
tex of the element complies with the associated plastic ow rule, the denition of p yields
then the ow rule holds at all points of the element domain. From !
Di X
nN
the previous statement, one can easily infer that the discrete con- i Ti i i Di i
p L xn BQ xn d  e 18
straint (12b) is perfectly equivalent to its continuous counterpart nN n1 nN
(6e). And, consequently, any coupling concerning the e variables is can-
In order to enforce the compatibility constraint (6c), a supple- celed out.
mentary quadratic penalty term, controlled by a scalar penalty
parameter r, is added to the Lagrangian functional: 2.3. Uzawa algorithm with block relaxation
r X
nE
Ld; e; l; lk L0 pTi pi 15 In this section, we outline the main steps of the Uzawa algorithm
2 i1
with block relaxation used to solve the Augmented Lagrangian form
26 A.N. Anto et al. / Computers and Geotechnics 41 (2012) 2335

of the discrete LA problem. As mentioned before, during the deriva- where nE and nN are the number of elements and the number of
tion of quadratic model, special attention was taken to ensure that nodes, respectively.
the minimization strategy employed for the original linear model Since STEPs 1 and 3 do not present any relevant differences
could be maintained without losing its effectiveness. A more de- from the implementation used for the linear model, we dispense
tailed explanation of the solution procedure can be found in [18], with any further description of this matter and instead focus our
complemented by the appendices of [17], with aspects concerning attention on STEP 2.
the implementation of commonly used yield criteria. The STEP 2 minimization depends signicantly on the yielding
In this text we focus our main attention on the implementation criterion used. The latter denes both the characteristics of the
issues that differ from the linear model. plastic dissipation function, D, and the normality space, Cc . We
The Uzawa algorithm with block relaxation comprises, as note that these are the two factors that determine the strategy
shown in Fig. 1, a cyclic two step minimization procedure followed and complexity of the minimization algorithm.
by a third stage where the Lagrange multipliers are updated. In the present work, the MohrCoulomb criteria is assumed,
Both the relaxation STEP 2 and the update of the Lagrange mul- which leads to a linear dissipation function and a non-linear nor-
tipliers (henceforth identied as STEP 3) consist of element-by-ele- mality rule constraint:
ment procedures. But the relaxation STEP 1 affects, simultaneously,
all the nodal velocities degrees of freedom, d, thus involving all the c0
De_ tre_ 27
elements contributions at the same time. The latter minimization tan /0
results in a system of linear equations with the following block e_ 2 Cc () tre_ P je_ I j je_ II j je_ III j sin / 28
structure, that must be solved for each iteration:
 ( )   In the previous expression, c0 and /0 represent the material effective
rA F dk f ek ; lk cohesion and friction angle, respectively. The strain rate compo-
19
FT 0 lk;k 1 nents are dened in the principal strain space.
Taking into account that the strain rate approximation (10b) is
where A is symmetric and positive denite. Thus, linear, the total element dissipation can be expressed as the sum of
!
1 F T yk  r the nodal contributions of the element vertices:
dk yk  b ; 20
r FT b Z X
nN
DLi ei dX Di Li xn ei 29
with Xi n1

yk A1 f ek ; lk 21a Moreover, if the normality rule holds at the vertices of the elements
bA F 1
21b it also holds throughout the elements domain.
As a consequence of the above statements, the minimization
In the governing system (19), all block denitions (A, F, f) can be de- algorithm recommended in [17] (see Appendix C) for the linear ele-
rived from the imposition of the stationarity condition ments formulation, which is performed directly with no iterations,
rLd; lk 0 22 can, in the particular case of MohrCoulomb material, also be
adopted for quadratic elements as long as it is applied vertex-by-
leading to vertex, separately.
XnE XnN
Di
A BQ i X n T BQ i X n 23 2.4. Parallel implementation
i1 n1
nN
XnE Z Z
The development of an efcient parallel implementation of our
F Q i T b dX Q i T t dC 24 model is greatly facilitated by the characteristics of the solution
i i
i1 X Cr
procedure adopted. Indeed, STEPs 2 and 3 being executed in an ele-
X nE X nN
Di
f ek ; lk e BQ i X n T rei i ment-by-element manner, their inclusion in a parallel scheme re-
F n  ln 25
i1 n1
nN quires no adjustments and displays perfect scalability features.
XnE Z Z Finally, STEP 1 requires the solution of a (often very large)
e
F ~ dX
Q i T b Q i T ~t dC 26 sparse system of linear Eq. (19). Fortunately, the governing system
i1 Xi Ci
r matrix remains unchanged throughout the entire iterative process,
and the factorization of the sub-matrix A, usually a signicant cost
in the overall computations, has to be performed just once.
In the last decades, a number of highly efcient direct parallel
solvers have been developed and are currently available to the sci-
entic community. In this work, we used one of those software
packages called PSPASES [22]. Our choice was based on the fact
that it is an open source package, developed for our kind of matri-
ces (symmetric, sparse, and positive denite) and, lastly, it allows a
good interface between the data structure of the nite element
code and the solver code.

3. The bearing capacity problem

As seen in Section 1, the bearing capacity of a strip footing acted


upon by a vertical centered load in a cohesionless material is given
by Eq. (2).
Assuming an associated ow rule, an exact [4] bearing capacity
Fig. 1. Uzawas block relaxation algorithm. factor Nq can be obtained:
A.N. Anto et al. / Computers and Geotechnics 41 (2012) 2335 27

 
/0 p tan /0 Table 2
Nq tan2 45 e 30 Common equations for shape factors sc and sq.
2
Proposal sc sq
Recently, two proposals for Nc have been made [23,24]. The values
Meyerhof [2] 1 + 0.1KpB/L 1 + 0.1KpB/L
obtained by [23] for rough footing can be accurately determined
De Beer [3] and Vesic [5] 1  0.4B/L 1 + (B/L) tan /0
using the following equation: Brinch Hansen [4] 1  0.4B/L 1 + (B/L) sin /0
p3p2 tan /0 2
Eurocode 7 [26] 1  0.3B/L 1 + (B/L) sin /0
Nc e 6 tan /0 5p 31
Kp = tan2 (45 + /0 /2).
and [24] used the method of characteristics to obtain exact values
for this bearing capacity. [25] proposed the following equation to
approximate the results obtained by [23]:

Nc Nq  1 tan1:32/0 32

Other common equations are the one from [2],

Nc Nq  1 tan1:4/0 33

and the one from [4],

Nc 1:5Nq  1 tan /0 34

Eurocode 7 [26] recommends the following equation for Nc:

Nc 2N q  1 tan /0 35

Fig. 2 shows the ratio between each proposal for the bearing capac-
ity factor Nc (for rough footing) and the exact values obtained by
Martin [24]. It can be seen that the solutions from [23,24] are prac-
tically the same for 15 6 /0 6 50 and that the equation proposed Fig. 3. Results obtained from the equations of sc and sq of Table 2 for /0 = 30.
by Salgado [25] conservatively estimates the results obtained by
Martin [24]. Below a friction angle of 1015 most proposals seem
to underestimate the bearing capacity factor, although its values equal to 1, whereas the other proposals give values less than or
are, for these friction angles, quite low and therefore the low ratio equal to 1. Also, Meyerhofs proposal for this shape factor is depen-
Nc/Nc, Martin is not signicant for practical purposes. It can also be dent on the soil friction angle, whereas the other proposals are not.
noted that the classic equations from [2,4] are appropriate for the
most commonly used values of the friction angle and the equation
from [26] clearly overestimates the bearing capacity factor. 4. Finite element upper-bound calculations
As also seen in Section 1, the three-dimensional effect can be ta-
ken into account by shape factors sc and sq. Common equations for 4.1. Denition of the numerical problem; results
sc and sq are summarized in Table 2.
Fig. 3 shows the comparison between the different equations of A set of three-dimensional nite element upper bound calcula-
sc and sq presented in Table 2, for the case /0 = 30. It can be seen tions using the nite element program SUBLIM3d [27] and the
that the different equations give results that can be quite different techniques described in Section 2 were performed to determine
for both shape factors sc and sq but particularly for the case of sc. In separately each of the two terms of (3) for different L/B ratios
fact, Meyerhofs proposal shows a shape factor sc greater than or and soil friction angles /0 .
All calculations assumed a rigid footing with a rough base. The
base of the footing was taken at the surface of the soil (Fig. 4) and a
surcharge q can be applied at the same level. Loads are centered
and vertical and therefore only a quarter of the problem was
modeled.
For the determination of the rst term of (3), qUB;BL
u;c , the sur-
charge q was considered equal to zero and a non-zero soil unit
weight was assumed. The different values of qUB;BL
u;c for different
L/B and /0 were obtained, which allowed [Ncsc]UB to be determined:

Fig. 2. Ratio between different proposals for Nc and the one from [24], for rough
footing base. Fig. 4. Geometry of the model.
28 A.N. Anto et al. / Computers and Geotechnics 41 (2012) 2335

Table 3
Values of [Ncsc]UB obtained for different L/B ratios and friction angles /0 .

L/B /0 = 15 20 25 30 35 37.5 40 42.5 45


1 1.0315 2.5977 6.5065 16.6006 45.0148 77.6057 136.4258 255.4178 490.4453
1.2 1.0912 2.7247 6.9575 16.9444 45.8705 77.9009 136.9326 252.1908 487.3068
1.5 1.1549 2.8262 6.7791 17.0236 45.4888 75.8355 131.7749 239.7184 460.5813
2 1.2411 3.0168 6.9809 17.5664 45.2404 74.9809 130.5883 235.1844 422.2497
3 1.2902 3.0747 7.2782 17.4747 43.7919 71.5015 120.9434 213.3200 383.9536
4 1.3169 3.1575 7.3277 17.4115 42.9310 69.2068 116.2855 206.0570 359.8267
5 1.3312 3.1800 7.3426 17.3226 42.2924 67.8038 113.7917 198.7584 350.1349

qUB;BL
u;c 4.2. Mechanisms
Nc sc UB 36
0:5cB
Fig. 6 shows the plastic deformation zones obtained for the cases
The results are shown in Table 3. of the determination of Ncsc and Nqsq for two values of the friction
Similarly, for obtaining the second term of (3), qUB;BL
u;q , the sur- angle25 and 40and two L/B ratios1.2 and 5. These plastic
charge q is assumed non-null and the soil unit weight c is consid- deformation zones correspond to zones with different values of
ered zero. Different calculations were performed using different L/ je_ I j je_ II j je_ III j. The plastic deformation zones allow the mecha-
B ratios and soil friction angles /0 . Values of [Nqsq]UB were deter- nisms involved in the failure to be inferred.
mined through Zones of more concentrated deformation can be observed in
qUB;BL this gure. This concentration is automatically dealt with by the
u;q
Nq sq UB 37 algorithm.
q
It can also be seen that the volumes involved in the mechanisms
The results are shown in Table 4. are greater for greater friction angles, and for greater L/B ratios.
The results presented in Tables 3 and 4 were obtained using They are also greater in the determination of Nqsq than in the deter-
about 800 thousand elements, which corresponds, more or less, mination of Ncsc.
to the available memory of the computer cluster used. This means Also, for /0 = 25 the width of the mechanism (which is approx-
that results have different levels of accuracy: The results of prob- imately 2B) for L/B = 5 and for the determination of Ncsc would not
lems with mechanisms involving lower volumes of soil (lower L/
B ratios and lower values of /0 ) will be more accurate than those
with mechanisms involving larger volumes. Table 5
Two examples of the nite element meshes used are presented Calculation times for four examples of calculations for L/B = 5.
in Fig. 5.
Case /0 () Number of global DOF Number of local DOF CPU time
Table 5 includes, for L/B = 5, the number of global degrees of
freedom, the number of the local degrees of freedom and the calcu- Ncsc 25 3,822,357 10,941,696 6h 13 min
Ncsc 40 3,910,781 11,191,680 8h 56 min
lation time for Ncsc and Nqsq and two values of the soil friction an-
Nqsq 25 4,070,493 11,643,264 5h 10 min
gle25 and 40. The general tendency illustrated by these Nqsq 40 3,993,493 11,425,536 9h 02 min
examples is that CPU time is greater for greated soil friction angles.

Table 4
Values of [Nqsq]UB obtained for different L/B ratios and friction angles /0 .

L/B /0 = 15 20 25 30 35 37.5 40 42.5 45


1 5.1655 9.4592 18.2830 36.6537 80.1535 123.0336 192.2343 319.0830 532.4616
1.2 5.1023 9.1299 17.4747 35.2508 77.2521 117.3832 186.5130 306.6485 511.8958
1.5 4.9451 8.9137 16.5252 33.2643 71.6946 107.5508 170.6644 277.8160 465.2778
2 4.7817 8.4899 15.4721 30.3298 64.5540 95.4602 150.0537 240.9592 402.7045
3 4.5933 7.9948 14.5333 26.9965 55.8177 82.3859 125.8146 199.4856 331.8542
4 4.4884 7.6326 13.6231 25.2171 50.9390 74.7811 113.5264 176.9865 291.0008
5 4.3787 7.3549 12.9892 24.1785 47.9338 69.9512 104.9227 162.7102 261.0559

Fig. 5. Examples of the nite element meshes used in the calculations. Each represented volume is subdivided into 24 ten-node tetrahedral nite elements. The two gures
have different scales.
A.N. Anto et al. / Computers and Geotechnics 41 (2012) 2335 29

increase signicantly if a greater L/B ratio was considered. This is for the case of the determination of Ncsc, L/B = 5 and /0 = 25. The
not the case either for /0 = 40 or for /0 = 25 and Nqsq. results plotted in this gure correspond to a coarser mesh than
For a better understanding of the mechanisms involved in the the ones used for obtaining the results previously presented, for
colapse of these structures the velocity eld is plotted in Fig. 7 visualization purposes.

Fig. 6. Plastic deformation zones for the determination of Ncsc and Nqsq for two values of the soil friction angle 25 and 40 and two L/B ratios 1.2 and 5.
30 A.N. Anto et al. / Computers and Geotechnics 41 (2012) 2335

4.3. Comparison with results obtained from other authors for comparison. They also include other results obtained within the
present work, marked as regular mesh which will be addressed
Figs. 8 and 9 include the results presented in Tables 3 and 4 for further in the paper.
/0 equal to 25, 30, 35, 40 and 45, as well as other results, used In Fig. 8 the following results are used for comparison:

 those obtained by Zhu and Michalowski [7] using the nite ele-
ment stressstrain code ABAQUS (except for /0 = 45, since
these authors do not provide this information); the values of
Ncsc are not directly reported by these authors; therefore, the
results presented in Fig. 8 correspond to the values of sc pre-
sented by the authors multiplied by the bearing capacity coef-
cient obtained by Michalowski [28], which should correspond
to the values effectively obtained by Zhu and Michalowski [7];
 those obtained by Lyamin et al. [8] using a nite element upper
and lower bound limit analysis and also average weighted val-
ues proposed by these authors;
 those obtained by Puzakov et al. [9] using the nite difference
code FLAC3D (except for /0 = 45, because those results are
not available from these authors); the values of Ncsc were not
Fig. 7. Velocity eld in impending colapse for the case of the determination of Ncsc,
L/B = 5 and /0 = 25.

Fig. 8. Values of Ncsc obtained from the UB nite element calculations: comparison with previously published results.
A.N. Anto et al. / Computers and Geotechnics 41 (2012) 2335 31

Fig. 9. Values of Nqsq obtained from the UB nite element calculations: comparison with previously published results.

Fig. 10. Comparison between the values of Nc and Nq obtained from the upper bound nite element calculations using two-dimensional regular meshes and the reference
values of [24] and from (30). The indicated values of L/B mean 2D regular meshes with the same discretization as the 3D ones that were used for each L/B.
32 A.N. Anto et al. / Computers and Geotechnics 41 (2012) 2335

given by these authors; the values represented in Fig. 8 corre- - use of Meyerhofs solution seems unsafe for very low L/B
spond to the values of Nc obtained by those authors using the ratios;
same code multiplied by the values of sc they report. - use of Nc,Martin(1  0.4B/L) is always safe;
- as remarked by other authors, [Ncsc] increases with the L/B
Also in Fig. 8 are two common solutions and one other, for ref- ratio for /0 < 30 and decreases with /B for /0 > 30, being
erence. The two common solutions are the ones from Eurocode 7 approximately constant for /0 = 30.
and from Meyerhof, using both Nc and sc suggested in [26,2]. The
other solution, as a reference, corresponds to the Nc from [23] The values of Nqsq obtained in the present work were presented
and the most common equation of sc, given by 1  0.4B/L. in Fig. 9, which also show the following results, for comparison:
Analysis of Fig. 8 allows the following remarks:
 those obtained by Michalowski [6], which are upper bound
 Firstly, comparison with other results issuing from limit analy- solutions based on three-dimensional mechanisms (the solu-
sis [8] shows that the previous upper bound solution is clearly tions by Lyamin et al. [8] are not directly comparable);
improved and that the gap with the best known lower bound  those obtained by Zhu and Michalowski [7], which uses the
solution is still quite signicant. Also, the results obtained are nite element stressstrain code ABAQUS; in fact the results
clearly above the average weighted values from these authors. for Nq sq were not directly obtained by these authors; they
 Comparison with other proposals show that: determined [Ncsc] and estimated sq using the approximate fol-
- there is reasonable agreement between the results obtained lowing equation, which would be exact if there were no tangen-
within the present work and those from [9], particularly for tial stresses applied by the footing to the soil:
/0 > 30; although the upper bound results obtained in the
present work are usually lower than those from [9]; the N c sc  tan /0 1
sq 38
exception seems to be for the case of /0 equal to 35 and Nc tan /0 1
40, for lower L/B ratios; The exact value of the bearing capacity factor Nc is known and it is
- use of Eurocode 7 seems to be unsafe for the cases of both used in this equation. The values represented in Fig. 9 are therefore
low values of /0 (25 and 30) and high L/B ratios; determined using Nq from (30) and sq from the previous equation.

Table 6
Values of sc obtained for different L/B ratios and friction angles /0 .

L/B /0 = 15 20 25 30 35 37.5 40 42.5 45


1 0.7725 0.8190 0.8961 1.0131 1.1880 1.2846 1.4360 1.6003 1.8432
1.2 0.8147 0.8621 0.9382 1.0425 1.2009 1.3104 1.4411 1.5898 1.8691
1.5 0.8527 0.8924 0.9492 1.0486 1.1821 1.2733 1.3909 1.5191 1.7440
2 0.8897 0.9252 0.9660 1.0430 1.1517 1.2309 1.3087 1.4295 1.5910
3 0.9234 0.9484 0.9759 1.0352 1.1001 1.1676 1.2464 1.3181 1.4467
4 0.9488 0.9630 0.9877 1.0170 1.0821 1.1281 1.1906 1.2519 1.3626
5 0.9595 0.9710 0.9895 1.0207 1.0676 1.0994 1.1424 1.2021 1.2783

Table 7
Values of sq obtained for different L/B ratios and friction angles /0 .

L/B /0 = 15 20 25 30 35 37.5 40 42.5 45


1 1.2825 1.4420 1.6577 1.9574 2.3483 2.6237 2.9284 3.4040 3.8988
1.2 1.2640 1.4090 1.6116 1.8820 2.2512 2.5026 2.8461 3.2666 3.9727
1.5 1.2233 1.3504 1.5224 1.7585 2.0625 2.2948 2.6343 2.9807 3.5844
2 1.1801 1.2845 1.4206 1.6064 1.8661 2.0813 2.2896 2.6798 3.0345
3 1.1300 1.2039 1.3004 1.4294 1.6215 1.7533 1.8903 2.1670 2.4444
4 1.1014 1.1564 1.2362 1.3633 1.4916 1.5615 1.6776 1.8258 2.0159
5 1.0799 1.1231 1.1799 1.2654 1.3850 1.4639 1.5746 1.6897 1.8344

Fig. 11. Values of sc and sq obtained from the upper bound nite element calculations using regular meshes.
A.N. Anto et al. / Computers and Geotechnics 41 (2012) 2335 33

Fig. 9 also includes three common solutions for Nqsq (in fact, for Nc sc UB
sc 39
sq, as the exact solution for Nq is known), using the three different Nc
equations for sq included in Table 2. Nq sq UB
Analysis of Fig. 9 allows the following remarks: sq 40
Nq
 the results obtained in the present work clearly improve the
other upper bound solution [6]; But which bearing capacity factors are appropriate?
 all common solutions lead to signicantly lower values of Nq sq One immediate answer to the previous question would be that
than the ones obtained from the calculations of the present the appropriate bearing capacity factor Nc is the one from [24] or
work; from [23] and, for the case of Nq, the one from (30). But it should
 all results obtained from [7] are lower than the ones obtained be taken into consideration that the results from Tables 3 and 4
from the calculations of the present work. are upper bounds and that those results are limited by the memory
capacity of the computer cluster, giving more accurate results for
5. Shape factors sc and sq the smaller problems and less accurate for the larger ones, as pre-
viously mentioned. This would mean that if the exact bearing
5.1. Determination of the shape factors capacity factors are used, the results from (39) and (40) would also
be upper bounds and therefore, for the cases of the problems with
Determination of the shape factors sc and sq can be performed less accurate nite element results, they could be signicantly on
using the results in Tables 3 and 4 and appropriate values of the the unsafe side. They would be, however, consistent, given the
bearing capacity factors Nc and Nq: scope of the methods being used in the present work.

Fig. 12. Values of sc obtained from the upper bound nite element calculations: comparison with previously published results.
34 A.N. Anto et al. / Computers and Geotechnics 41 (2012) 2335

But what would be the appropriate way of obtaining more ade- This gure shows that the results for Nq are much more accurate
quate (in practical terms) results for the shape factors with this than the ones for Nc. This probably means that the results for Nqsq
tool? Such results could be obtained if the bearing capacity factors are more accurate than those for Ncsc. And the gure also shows
used in (39) and (40) were similarly unsafe. This could be found the importance of the use of the procedure for determining the
using the same types of mesh used in each 3D calculation in 2D shape factors previously described: it can be seen that the results
plane strain calculations for determining the bearing capacity for the bearing capacity factors are relatively dependent on the -
factors. However, this is not possible to do exactly as described be- nite element mesh being used.
cause of the fact that the nite element meshes were prepared to The shape factors were determined using
obtain the best possible results with the available memory and,
therefore, are not regular.
Nc sc UB
regular mesh
All computations for the determination of [Ncsc]UB and [Nqsq]UB sc 41
were therefore repeated using less accurate but regular nite ele- NUB
c;regular mesh
ment meshes. Being upper bounds, less accurate means that greater Nq sq UB
regular mesh
collapse loads are obtained, as can be seen in Figs. 8 and 9. Two- sq 42
NUB
q;regular mesh
dimensional nite element meshes were used to estimate, for each
different mesh used, the bearing capacity factors that could be The results of the shape factors obtained using these equations
achieved; these meshes were different for different L/B ratios and are shown in Tables 6 and 7 and represented in Fig. 11. It is clear that
for different friction angles. The results of Nc and Nq obtained from both the sc and sq shape factors depend on the soil friction angle. The
the nite element calculations are, in Fig. 10, compared respectively shape factor sq is always greater than 1, as in all known proposals.
with the values from [24] and from (30). The shape factor sc is less than or equal to 1 for friction angles less

Fig. 13. Values of sq obtained from the upper bound nite element calculations: comparison with previously published results.
A.N. Anto et al. / Computers and Geotechnics 41 (2012) 2335 35

than 30; it is approximately 1 (almost independently of L/B) for /0 published results of the same type, but they should be, from a prac-
equal to 30 and greater than or equal to 1 for greater friction angles. tical point of view, unsafe.
The values of sq are much greater than the values of sc. Towards a more practical approach, the results of Ncsc and Nqsq
The values of sc and sq determined using (41) and (42) and were divided by Nc and Nq determined with comparable (upper
shown in Tables 6 and 7 and in Fig. 11 are approximations of the bound) accuracy. These shape factors are proposed in the present
shape factors that should be used with exact values of Nc and Nq. work, for use with the exact values of Nc and Nq.
It should be noted that the proposed shape factor sq was not tested
for use in combination with a depth factor taking into account the Acknowledgement
resistance of the soil above the foundation base.
The authors thank the nancial contributions given to this pa-
5.2. Comparison with results from other authors per by Fundao para a Cincia e Tecnologia (Project PTDC/ECM/
70368/2006).
The results of the obtained values of the obtained shape factors
sc and sq are represented in Figs. 12 and 13, for ve values of the References
soil friction angle. It can be seen that, as expected, the values of
the shape factor using the regular meshes are considerably more [1] Terzaghi K. Theoretical soil mechanics. 2nd ed. New York: John Wiley and
Sons; 1943. 510 pp.
conservative than the ones obtained from the best results. This is [2] Meyerhof GG. Some recent research on bearing capacity of foundations. Can
particularly true for the case of sc. Geotech J 1963;1(1):1626.
It should also be emphasized that these shape factors are the [3] De Beer EE. Experimental determination of the shape factors and the bearing
capacity factors of sand. Gotechnique 1970;20:387411.
best estimates that could be found using the available tools. But [4] Brinch Hansen J. A revised and extended formula for bearing capacity. Bull
they are no longer guaranteed to be upper bounds. Danish Geotech Inst 1970;28:511.
The results for sc obtained using the regular meshes and the [5] Vesic AS. Analysis of ultimate loads of shallow foundations. ASCE J Soil Mech
Found Div 1973;99(1):4573.
procedure described above show good agreement with the results [6] Michalowski RL. Upper-bound load estimates on square and rectangular
obtained by Zhu and Michalowski [7] and by Puzakov et al. [9], for footings. Gotechnique 2001;51(9):78798.
L/B > 2. They are greater than the values obtained by those authors, [7] Zhu M, Michalowski RL. Shape factors for limit loads on square and rectangular
footings. ASCE J Geotech Geoenviron Eng 2005;131(2):22331.
particularly for L/B < 2. The values obtained using the regular
[8] Lyamin AV, Salgado R, Sloan SW, Prezzi M. Two- and three-dimensional
meshes are close to the ones proposed by Meyerhof [2], for /0 = bearing capacity of footings in sand. Gotechnique 2007;47(8):64762.
40; the agreement is not as good for other values of the friction [9] Puzakov V, Drescher A, Michalowski RL. Shape factor sc for shallow footings.
Geomech Eng 2009;1(2):11320.
angle.
[10] Davis R, Selvadurai A. Plasticity and geomechanics. Cambridge University
The results for sc obtained using the best Ncsc and the exact val- Press; 2002.
ues of Nc (the ones that are guaranteed to be upper bounds) clearly [11] Lubliner J. The theory of plasticity. New York: Macmillan Publishing Company;
improve the previous upper bound solution [8]. 1990.
[12] Christiansen E. Limit analysis of collapse states. In: Ciarlet P, Lions J, editors.
The results for sq obtained using the regular meshes are in Handbook of numerical analysis (Part 2), vol. IV. North-Holland; 1996. p.
agreement with the ones from [7] for L/B P 3. The results from 193312.
these authors give lower values of sq, particularly for L/B < 3. The [13] Krabbenhft K, Lyamin A, Sloan S. Formulation and solution of some plasticity
problems as conic programs. Int J Solids Struct 2007;44:153349.
obtained results for sq using the regular meshes can be directly [14] Makrodimopoulos A, Martin C. Upper bound limit analysis using simplex
compared with the values from [6], who used a similar procedure strain elements and second-order cone programming. Int J Numer Anal
to estimate the shape factors. The results from these authors Methods Geomech 2007;31:83565.
[15] Krabbenhft K, Lyamin A, Sloan S. Three-dimensional Mohr-coulomb limit
roughly agree with the obtained in the present work for the greater analysis using semidenite programming. Commun Numer Methods Eng
values of L/B; for lesser values of this ratio, the results obtained in 2008;24(11):110719. doi:10.1002/cnm.1018.
the present work are considerably less than the ones from that [16] Franck Pastor JP, Loute Etienne. Limit analysis and convex programming: a
decomposition approach of the kinematic mixed method. Int J Numer Methods
author. The common equations for the shape factor sq seem conser-
Eng 2009;78(3):25474.
vative for all cases. [17] Vicente da Silva M, Anto AN. A non-linear programming method approach for
upper bound limit analysis. Int J Numer Methods Eng 2007;72:1192218.
doi:10.1002/nme.2061.
6. Conclusions [18] Vicente da Silva M, Anto A. Upper bound limit analysis with a parallel mixed
nite element formulation. Int J Solids Struct 2008;45(2223):5788804.
Determination of the shape factors for use in the bearing capac- doi:10.1016/j.ijsolstr.2008.06.012.
[19] Zienkiewicz O, Taylor R. The nite element method. Solid and uid mechanics:
ity equation is a traditional foundation engineering problem. The dynamics and non-linearity, 4th ed., vol. 2. McGraw-Hill, Inc.; 1997.
most common proposals for such shape coefcients give quite dif- [20] Zienkiewicz O, Taylor R. The nite element method. Basic formulation and
ferent results. A set of three-dimensional nite element upper linear problems, 4th ed., vol. 1. McGraw-Hill, Inc.; 1997.
[21] Vicente da Silva M. Implementao numrica tridimensional do teorema
bound calculations were performed to determine the bearing cinemtico da anlise limite. Ph.D. thesis, Universidade Nova de Lisboa;
capacity of rigid footings with a rough base, for different soil fric- 2009.
tion angles and different footing rectangular ratios. [22] Joshi M, Karypis G, Kumar V, Gupta A, Gustavson F. Pspases: an efcient and
scalable parallel sparse direct solver. In: Proceedings of the ninth SIAM
For these calculations, a nite element formulation was devel-
conference on parallel processing for scientic computing; 1999.
oped, based on a mixed approach with quadratic approximation [23] Hjiaj M, Lyamin AV, Sloan SW. Numerical limit analysis solutions for the
for the velocities and linear approximation for the plastic strains, bearing capacity factor Nc. Int J Solids Struct 2005;42:1681704.
implemented in a parallel computing framework. [24] Martin CM. Exact bearing capacity calculations using the method of
characteristics. In: Proc. 11th int. conf. IACMAG, vol. 4, Turin; 2005. p. 441-450.
The obtained bearing capacity was compared with ones from [25] Salgado R. The engineering of foundations. McGraw-Hill; 2008.
other authors and, in particular, they show a signicant improve [26] Eurocode 7. Geotechnical design, part 1: general rules. CEN, European
on the results of the best available upper bound solutions. Committee for Standardization, Brussels; 2004.
[27] Vicente da Silva M. SUBLIM3d strict upper bound limit analysis code
The obtained values of Ncsc and Nqsq were divided by the exact homepage; March 2011. <http://www.dec.fct.unl.pt/projectos/SUBLIM3d/>.
solutions for Nc and Nq. Therefore, strict upper bound approxima- [28] Michalowski RL. An estimate of the inuence of soil weight on bearing
tions of the shape factors were obtained. The results of these capacity using limit analysis. Soils Found 1997;37(4):5764.
upper-bound shape factors show a considerable improvement over

Вам также может понравиться