You are on page 1of 6

Yield loci for shallow foundations by “swipe” testing

G. Gottardi & L. Govoni


University of Bologna, Bologna, Italy
R.Butterfield
University of Southampton, Southampton, UK

ABSTRACT: It is now well established that the load-displacement response of a shallow foundation can be
modelled as a work-hardening system in a three-dimensional (V,M/B,H) load space. Nevertheless,
establishing the form of a representative set of yield-loci from physical experiments, using load-controlled
tests, is a formidable task. The paper presents results from, and analytically justifies, a displacement
controlled ‘swipe’ testing technique that generates a complete yield locus from a single test. Such tests are
simple to simulate numerically and results are shown which closely replicate the experimentally determined
yield loci for surface footings. Preliminary results are also presented from an extension of the numerical
‘swipe’ test modelling to investigate buried footings.

or h / h0  m / m0  4v1  v  (1b)
1 INTERACTION DIAGRAMS FOR where: h = H/Vmax, m = M/BVmax, v = V/Vmax and h0
PREDICTING LOAD CAPACITY OF = Hmax/Vmax and m0 = Mmax/BVmax.
FOOTINGS In these dimensionless equations Vmax is the
vertical, central-load capacity of the footing, which
The shortcomings of conventional bearing capacity therefore automatically takes account of its shape.
analyses based on Brinch Hansen’s well-known The values of (m0, h0), established from these
equation, (i.e the need for the fundamental ‘centre- tests and from very many more sophisticated ones
line load’ capacity to be modified by empirical on dense sand (Ticof 1977, Gottardi 1992, Gottardi
multiplication factors for each of its terms to allow & Butterfield 1993), loose sand (Nova & Montrasio
for footing shape, load inclination, eccentricity and 1991) and a Kaolin clay (Martin & Houlsby, 2000),
so on) led to a proposal (Butterfield & Ticof 1979; are consistently about (0.09, 0.123).
Butterfield 1980) that a better solution might be The form of the full three-dimensional surface
possible using the concept of an interaction diagram (sketched in Figure 3) was explored by Butterfield &
- a well-established procedure in structural Gottardi (1994). Longitudinal sections of this figure
engineering. are parabolas and transverse sections geometrically
The idea, applied initially to shallow foundations similar, inclined ellipses.
on dense sand, was that all load combinations The equation to the complete surface is:
leading to failure of such a foundation, when plotted
in a three-dimensional load space (V,M/B,H) - i.e. m / m0 2   h / h0 2  2 amh  4v1  v 2  0 (2)
vertical, moment/footing-breadth and horizontal m0.h 0
loads – will define a ‘failure surface’ for a specific
soil-footing system (Butterfield 1980). in which a = -0.22, corresponding to a 12.7 degree
Early experiments, using footings with different rotation of the ellipse.
aspect ratios, bearing on dense sand and brass-rod- It is interesting to note that essentially identical
model material (Butterfield 1981), established that parabolic and elliptical yield loci have been obtained
both the (V,H) and (V,M/B) sections of this surface (Zdravkovic et al. 2002) from finite element
were remarkably close to simple parabolas of the modelling of strip footings on sand.
form,
H / H max  M / M max  4V / Vmax 1  V / Vmax  (1a)

1
2 DEVELOPMENT OF A WORK-HARDENING This is clearly a very arduous process in which a
MODEL sophisticated load-controlled test and a laboriously
prepared soil bed are needed in order to establish a
Since the failure-load points for footings on dense single point on a yield-locus passing through the
sand identified accurately the onset of a work- selected value of V0.
softening process, it was also hypothesised
(Butterfield 1980, 1981) that all such points might
3.2 Displacement controlled tests
lie on the outermost yield-surface of a work-
hardening system and, furthermore, that the yield- Fortunately, there is an alternative method,
surfaces of the system might be a family of whereby a single displacement-controlled test (a
geometrically similar figures with the maximum ‘swipe’ test) can generate a yield locus, embedded in
previous vertical centreline-load serving as the the 3D yield-surface corresponding to any value of
hardening parameter (i.e. that the yield surfaces V0. The idea of a ‘side-swipe’ test was first mooted
might be a nested set of surfaces described by by Tan (1990) based on an analogy between (a) an
suitably scaled versions of equations (1) and (2)). undrained (zero volume change) triaxial test on a
The key ingredients of such a model are (Nova & saturated soil sample, consolidated to p0’, say,
Montrasio 1991): (a) the nested family of yield– compressed axially to failure and interpreted in a
surfaces; (b) the work-hardening process (e.g. by CSM framework and (b) a horizontal-displacement
using the results of a vertical centreline-load test, w test to failure of a footing (preloaded to V0) under
versus V – analogously to the role of an e-p’ curve in conditions of zero vertical displacement and zero
Critical State soil models) and (c) a set of plastic rotation.
potentials. A successful model of this kind would Tan concluded that, if the elastic displacements in
enable plastic displacement increments, vertical, both tests were essentially zero, the load–paths
rotational and horizontal (δwp, δBθp, δup), generated in the tests would trace the associated
corresponding to load increments (δV,δM/B,δH), to yield-loci. He was, in fact, using the previously
be predicted for any loading programme applied to mentioned analogy between “specific volume versus
the footing. The total displacement increments are p’ ” work-hardening in CSM and “w versus V”
then the sum of the plastic and the associated elastic work-hardening for a centrally loaded footing.
components (e.g. for the vertical displacement, δw = Figure 2 illustrates a ‘side-swipe’ test which
δwe + δwp). generates an approximate yield locus for a soil-
footing system.
Martin & Houlsby (2000) extended the analogy to
3 EXPERIMENTAL DETERMINATION OF encompass a rotation-controlled ‘swipe’ test; one
YIELD-LOCI carried out, from a specific V0 value, under
conditions of zero vertical and horizontal
displacement. A general ‘swipe’ test on a footing
3.1 Load controlled tests
(Gottardi et al. 1999) then becomes one in which no
Identification of yield-loci lying on the (V,M/B,H) vertical displacement is allowed (w = w0 = constant)
yield surface is clearly a task of paramount but any specified (Bθ, u) displacement regime may
importance in validating such models. be imposed throughout the test (Figure 3).
One method of doing so, used by Gottardi (1992) A method of establishing a formally correct
is to load the footing to V0, unload it to a lower yield–locus from any swipe test is presented in
vertical load followed by reloading to yield along a Section 4.
selected load-path. For example, Figure 1 shows a
number of his load-paths in the (V/V0-H/V0) plane.
H H
0.2

H/V0

V u
M2 A3
0.1 G2
D4
D3
X D1 V u

B6 Y Z

0.0 w w
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

V/V0
Figure 1. Experimental load-paths in load-controlled tests
(from Gottardi & Butterfield 1993). Figure 2. Output from a horizontal ‘swipe’ test on a footing.

2
3.3 Test results 0, which is derived from equation (2) (Butterfield
and Gottardi, 1996) by defining q as,
The validity of the swipe-testing process was
investigated by Gottardi et al (1999) using a 100 mm m2 h2 hm
diameter, rigid, circular footing on dense sand. q 2
 2
 2a (3)
m0 h0 h0 m0
The key results from this work were that, for
footings preloaded to a vertical load V0: (b) Alternatively, by dividing equation (2)
(a) There was indeed a family of nested yield- throughout by 4v 1  v 2 and defining (mn, hn) = (m,
surfaces of the same form as equations (1) and (2), h)/[4v(1-v)], all points on the ‘cigar’ can be reduced
with (V0, H0) replacing (Vmax, Hmax) in (h, m, h0, m0). to a single, central ellipse (Gottardi et al. 1999).
The values of (m0, h0, a0), obtained from a Figure 5 shows their data plotted this way, together
statistically best-fitted ‘cigar’ to the swipe- with the best-fit ellipse. Points inside the ellipse
determined yield–loci, were (0.09, 0.121, -0.223), belong to elastic reloading paths for footings which
values essentially identical to those that described had been loaded to V0 and partially unloaded before
the failure envelope. Figure 4 shows a normalised swiping to yield (see Section 5).
plot of a set of parabolic yield-loci in the (v, q) (c) The simple, one-degree-of-freedom,
plane. hardening model could provide reasonable
predictions of plastic displacement increments.
(d) The ‘flow’ was approximately associated in
H (M/B,H) planes (i.e. the plastic potential and the
yield surface could be assumed identical), but this
was not so for load-paths along which V did not
remain constant. An adequate plastic potential is still
to be defined for such cases.
M/2R
hn
0.15
Best fit
V GG03
0.10 GG04

Swipe paths GG05


GG06
0.05
Figure 3. Sketch of load paths followed in ‘swipe’ tests (from GG28

Gottardi et al. 1999). GG29

0.00 mn GG07
-0.15 -0.10 -0.05 0.00 0.05 0.10 0.15
GG08

1.2 GG10
-0.05
q GG12

1.0 -0.10

-0.15
0.8

Figure 5. Dimensionless elliptical section of yield surface


(from Gottardi et al. 1999).
0.6

0.4
4 JUSTIFICATION AND INTERPRETATION OF
‘SWIPE’ TESTING

0.2 If, in a classical work-hardening plasticity model of


the soil-footing system, hardening were controlled
solely by V0, which in general might be a function of
0.0 the three plastic displacement components (wp, Bθp,
0.0 0.2 0.4 0.6 0.8 v 1.0
up), then V0 would change according to:
Figure 4. Dimensionless parabolic sections of yield surface. V0 V0 V0
V0  w p   p  u p (4)
w p
 p
u p
These cover a range of swipe tests (e.g.
increasing u only; increasing θ only and increasing In the model being considered hardening only occurs
both u and θ in a fixed ratio) for V0 = 4/5 Vmax. The as a result of increasing δwp. Hence, to avoid
normalised parabolas have the equation q - 4v(1-v) = hardening, and thereby, ensure no change of yield-

3
locus, any displacement path imposed on the footing geometrically similar then the succession of points
must ensure that δwp = 0 (i.e. that the vertical plastic {…(H1/V01, V1/V01), (H2/V02,V2/V02), ….} will trace
displacement component does not change during the out the generic yield locus in a dimensionless (H, V)
process). Since, in a swipe test, the footing is driven plane.
along an arbitrary, extensive horizontal and/or Figure 7 shows the output of a horizontal swipe
rotational displacement path this will necessarily test corrected this way.
impose plastic deformations – there can be no Figure 7
1
question of such a process being one of simple
H
locus 2
elastic-unloading, even though the vertical load may 0.8

H2 locus 1 b2
decrease. 0.6
'swipe' test
The differential form of a yield surface load-path
f(V,M,H,V0) is, 0.4
H1
b1

f f f f
0.2

V  M  H  V0  0 (5) V2 V1
V M H V0 0
V01 V02
0.6 0.7 0.8 V 0.9 1

Since a perfect swipe test ensures δV0 = 0, then


equation (5) requires the two vectors, w = w0 a1

 f f f  a2
 , ,  and V , M , H  (6)
 V M H  w

to be orthogonal. The first of these is normal to f


and, therefore, the second, the load increment vector, Figure 6. Correction procedure for a horizontal swipe test.
is locally tangential to f, i.e. the load path in any
perfect swipe test necessarily traces a curve 0.20

embedded in the three-dimensional yield-surface h=H/Vo Uncorrected

corresponding to V0 (Butterfield & Gottardi 2003). Corrected for rig flexibility

Unfortunately, from an experimental point of 0.15 Corrected for rig flexibility and soil elasticity

view, it is very much simpler to impose the


condition δw = 0 rather than the formal requirement
δwp = 0. Therefore, due to the decrease in V, which 0.10

occurs in a swipe test, the associated elastic vertical


displacement recovery of the soil means that δwp ≠
0. A similar error can develop in a practical swipe 0.05
test due to the elastic deformation of the loading
frame. In their experiments using dense sand,
Gottardi et al. (1999) found the latter to be much the 0.00
more significant source of error. Both of these 0.00 0.20 0.40 0.60 0.80
v=V/Vo
1.00

errors, if small, can be corrected if reasonable values


of the vertical elastic compressibility of the soil- Figure 7. Swipe test output (w = constant) corrected to provide
a wp = constant locus.
footing system and the loading frame are known.
The remaining source of error, which cannot be
corrected directly, is that the hardening parameter V0
5 NUMERICAL MODELLING
may depend on (δθp, δup) in addition to wp. Unless
such dependence is also small, as it appears to be
Even with the improved efficiency of yield–curve
from the results of Gottardi et al. (1999), the concept
determination provided by swipe testing, it is still a
of δwp = 0 swipe testing is clearly no longer valid.
formidable experimental task to verify a three-
The ‘correction’ procedure for a δw = 0
dimensional yield surface in an adequate way,
horizontal swipe test is illustrated in Figure 6. At
especially when investigating embedded footings
any stage in the test, the load point b1 will lie on a
(Bransby & Randolph, 1999). One possible
yield locus, defined here by V01. In the ‘hardening
approach is to use numerical modelling for this
plane’ it will lie on an elastic unloading line at a1,
purpose. In principle, swipe tests can be performed
with w = w0, say. A small horizontal swipe
with wp = constant and the problems of soil bed
increment generates a load path increment b1 to b2
replication and rig stiffness eliminated. Constructing
and a2 will be the point in the V – w plane
sets of yield loci this way would be particularly
corresponding to b2 (since δw = 0 throughout the
convenient.
test). If the unloading line through a2 intersects the
A preliminary numerical investigation, using a
‘virgin’ V – w curve at V02 then b2 lies on the yield
2D finite-difference code (ITASCA, 2002), was
locus corresponding to V02. If all the yield loci are

4
carried out to establish whether the geometrical form footing paths from rotational swiping proceed only
of the curves already determined experimentally in a clockwise direction, those from horizontal
could be replicated. The soil bed (18 m wide x 8 m swiping move in either direction.
deep) was modelled as a standard elastic, perfectly Figure 12 shows the Gottardi et al (1999)
plastic, Mohr-Coulomb frictional material (φ = 30°; horizontal and rotational swipe data from their
ψ = 15°) and the footing (breadth B = 2 m) as a rigid experiments, which is clearly very similar to the
structural beam, initially laying on top of it, with a computer model output.
fully rough interface (Figure 8). The buried footing swipe tests also generated
essentially parabolic load paths. Embedment of the
footing caused not only an approximate doubling of
w its vertical load capacity (Figure 9), but also a small
u q
expansion of its (V/V0,H/V0) yield envelope (from
 about V0/8 to V0/7) and, in the case of the physically
buried footing, a small lateral shift towards the
origin of its peak load point. The (V/V0,M/BV0)
envelope also expanded, but its peak load point
shifted slightly away from the load origin.

1800

Load V [KN]
1600
w = applied constant vertical rate 1400
u = applied constant horizontal rate
 = applied constant rotation
1200
Figure 8. Geometry of the finite difference numerical model. 1000

800
A buried footing was then modelled in two ways:
(1) a 1 m deep soil surcharge (i.e. 17.6 kPa) was 600 Surface footing
added before displacements start and (2) the footing 400 Surface footing with surcharge
was physically buried at 1 m depth (D/B = 0.5) and 200
Embedded footing
the bed depth increased to 9 m. Load capacity according toTerzaghi
Horizontal and rotational swipe tests were 0
0 0.05 0.1 0.15 0.2 0.25
performed as follows. In all tests a vertical Displacement w [m]
downward movement sufficient to impose a specific Figure 9. Vertical load (V) versus displacement (w) curves.
V0, 80% of Vmax, was applied; in some cases the
footing was unloaded to about V0/8 before swipe H/Vo
testing (preloaded footings). In all tests swiping was 0.16

continued until a steady state, at which all loads 0.14


remained constant, was achieved. 0.12
The vertical load test curves for the three cases 0.1
mentioned (Figure 9) show that: values of Vmax are 0.08
very close to the Terzaghi predicted values; there is Surface footing
0.06
Surface footing with surcharge
no work softening (since the M/C model does not 0.04
allow it); the difference between the first loading Embedded footing
0.02
and unload-reload slopes is much smaller than those 0
measured in experiments. Since the M/C model has 0 0.2 0.4 0.6 0.8 1 V/Vo
not generated wp = constant loci the following Figure 10. Horizontal swipe test load paths.
comparisons relate only to the form of the swipe
curves themselves. M/BVo
The load paths for the three horizontal swipe tests 0.16
are shown in Figure 10 (each curve has been 0.14
normalized by the V0 value specific to each test). 0.12
Results for both normally loaded and preloaded 0.1
footing tests converge satisfactorily.
0.08
In the rotational swipe tests a tensile bond Surface footing
0.06
strength of 10 kPa was included at the footing-soil
0.04 Surface footing with surcharge
interface. The bonded and unbonded swipe curves
(not shown) were almost identical. 0.02 Embedded footing

Figure 11 shows the output from these tests and it 0


0 0.2 0.4 0.6 0.8 1
V/Vo
is interesting to note that whereas the preloaded
Figure 11. Rotational swipe test load paths.

5
250
rotational swipe tests with w = constant. The
H (N) analysis is currently being refined to perform
200
GG03 ‘perfect’ swipe tests throughout which wp =
constant, in order to provide yield loci directly.
150 Scale effects, if any, between model and
prototype foundations have yet to be fully
100 determined, although, since they will be reflected in
GG07
the centreline-load capacity of the system, the
50 dimensionless equations to the yield loci from model
tests may still be applicable. Large-scale or
0 centrifuge rotational swipe testing may well be a
0 200 400 600 800 1000 1200 1400 1600
V (N)
practical proposition.
200

M/B (N)
GG04
150 REFERENCES
GG08

100
Bransby M.F., Randolph, M.F 1999. The effect of embedment
depth on the undrained responde skirted foundations.to
combined loading. Soils and foundations 39(4), 19-33.
50 Butterfield R. 1980. A simple analysis of the load capacity of
rigid footings on granular materials. Journée de
0 Géotechnique, ENTPE, Lyon, France, 128-137.
0 200 400 600 800 1000 1200 1400 1600 Butterfield R. 1981. Another look at gravity platform
V (N) foundations. SMFE in Offshore Technology, CISM, Udine,
Figure 12. Experimental data, horizontal and rotational swipe Italy.
tests on surface footing (from Gottardi et al. 1999). Butterfield R., Gottardi G. 1994. A complete three–
dimensional failure envelope for shallow footings on sand.
Géotechnique, 44(1), 181-184.
Butterfield R., Gottardi G. 1996. Simplified failure envelopes
6 FINAL REMARKS for shallow foundations on dense sand. Int. Jour. Offshore
and Polar Engineering, 6(1),58-64.
There is now a considerable body of evidence that Butterfield R., Gottardi G. 2003. Determination of yield curves
the response of shallow, rigid foundations to either for shallow foundations by “swipe” testing. Proc. 1st
changes (δV, δM/B, δH) in the applied load or International symposium on shallow foundation,
changes in the associated displacements (δw, δBθ, FONDSUP, Paris, France, 1, 111-118.
Butterfield R., Ticof J. 1979. The use of physical models in
δu) can be modelled, with reasonable accuracy, as a design (discussion). Proc.7th ECSMFE, Brighton, UK, 4,
work-hardening plastic system with a single 259-261.
hardening parameter V0 dependent only on the Gottardi G. 1992. Modellazione del comportamento di
vertical plastic displacement component wp. fondazioni superficiali su sabbia soggette a diverse
The determination of yield-loci (defining in condizioni di carico. Ph.D. Thesis, University of Padova,
Italy.
aggregate a three-dimensional yield-surface for the Gottardi G., Butterfield R. 1993. On the bearing capacity of
foundation-soil system) from experiments is a surface footings on sand under general planar loads. Soils
crucial part of the validation of such models. A and Foundations, 33(3), 68-79.
technique for doing this efficiently, a displacement- Gottardi G., Houlsby G.T., Butterfield R. 1999. The plastic
controlled ‘swipe’ test, has been described and, response of circular footings on sand under general planar
when corrected for soil compressibility, formally loading. Géotechnique,49(4),453-469.
ITASCA Consulting Group (2002) FLAC: Fast Lagrangian
shown to generate the correct yield locus, if soil Analysis of Continua. Minneapolis, USA.
hardening is governed solely by the plastic Martin C.M., Houlsby G.T. 2000. Combined loading of
component of the V – w curve for the system. The spudacan foundations on clay: laboratory test.
results obtained replicated very closely those Géotechnique 50(4), 325-338.
obtained from vastly more laborious load-controlled Nova R., Montrasio L. 1991. Settlements of shallow
foundations on sand. Géotechnique 41(2), 243-256.
tests. Tan K. 1990. Centrifuge and theoretical modelling of footings
A preliminary numerical investigation, using a on sand. Ph.D. Thesis, University of Cambridge, UK.
well established constitutive model, has been able to Ticof J. 1977. Surface footings on sand under general planar
reproduce the key features of experimental results loads. Ph.D. Thesis, University of Southampton, UK.
for surface footings with surprising accuracy. Zdravkovic L., Ng P. M. and Potts D. M. 2002. Bearing
Encouraged by these results, the analysis was capacity of surface footings on sand subjected to combined
loading. Proc. Int, Conf. on Num. Methods in Geotech.
extended to include embedded footings (D/B = 0.5). Eng., NUMGE V, 323-330.
The dimensionless load paths generated were again
of parabolic form with, as expected, peak values
increasing to around 0.14 for both horizontal and

You might also like