Geotechnical and Geophysical Site Characterization – Huang & Mayne (eds)

© 2008 Taylor & Francis Group, London, ISBN 978-0-415-46936-4

SBPM testing in Bothkennar clay: structure effects

M. Arroyo, N. González, J. Butlanska & A. Gens

Technical University of Catalunya, (UPC), Barcelona, Spain

J.C.P. Dalton
Cambridge InSitu Ltd., Little Eversden, UK

ABSTRACT: A series of self-boring pressuremeter tests performed in Bothkennar clay are simulated using a
finite element model of undrained cavity expansion. A bonded elasto plastic soil model is employed to repre-
sent Bothkennar clay behaviour. The model is calibrated using laboratory test results. Then, the pressuremeter
expansion curve and the cavity pore pressure curves are predicted, both for tests made at depths where laboratory
results are available and for tests where such information is missing. Comparison is made in terms of the whole
response curves and of standard field test interpretation. The effect of structure is made explicitly visible by
comparing simulations with and without structure. Introducing structure in the model considerably improves the
simulation results. This means that SBPM tests seem able to detect most of the original soil structure.

1 INTRODUCTION conditions. This makes it a good initial candidate for

structure determination.
The inclusion of a measure of structure and bonding in Gonzalez et al. (2007) presented simulations of
modified elasto-plastic soil models is now widespread. cylindrical cavity expansion in clay using a bonded
The motivation behind this development is the consis- elasto-plastic model where the effect of structure was
tent laboratory observation of important behavioural explicitly included. The effect of initial structure was
differences, when subjected to similar loading paths, made patent in the simulations through a sensitivity
between intact and reconstituted soils (e.g. Cotecchia analysis. Here that approach is taken one step further
& Chandler, 2000). Such differences appear notwith- and an attempt is made to simulate a series of real
standing a common composition, porosity and stress SBPM tests in Bothkennar clay. The fact that Bothken-
state, and therefore elevate the variable(s) embody- nar clay has been extensively tested in the geotechnical
ing structure and/or bonding to the category of basic laboratory allows us to use laboratory data to predict
state variables, on an equal footing with stress and the SBPM test response.
overconsolidation ratio. In the next section the modelling approach
Elasto-plastic bonded soil models have proved their employed is presented. The SBPM data employed
abilities in numerical simulations of laboratory tests is described and the model calibration procedures
(e.g. for clays, Rouainia & Wood, 2000; Baudet & detailed afterwards. Various simulation results are then
Stallebrass, 2004). Their implementation into finite illustrated before drawing some conclusions.
element codes is also well advanced. One basic obsta-
cle, however, hinders their general application to engi- 2 MODELLING APPROACH
neering problems, and that is the difficulty associated
with the determination of structure. The essential aspects of the modelling approach fol-
Laboratory procedures for this purpose are rela- lowed are now described. More details can be found
tively cumbersome and require high quality samples. in Gonzalez et al. (2007).
A possible alternative is the employment of in-situ
measurements to retrieve soil structure parameters.
2.1 Material model
Soil structure is fragile. Hence, the measurement
methods most suitable for its determination are those The bonded elasto-plastic model here employed is
which involve minimal soil disturbance. The self bor- based on the CASM (Clay and Sand Model) developed
ing pressuremeter (SBPM) is widely regarded as the by Yu (1998). The CASM has been here both simpli-
mechanical in-situ test that best preserves undisturbed fied and extended. The main simplification introduced

257
is that of assuming associated behaviour. The main
extension is that of introducing a new scalar history
variable, b, representing “bonding”.
The way this bonding variable enters the model fol-
lows closely the original proposal of Gens & Nova
(1993). The yield surface is assumed to enlarge with
increasing amount of bonding in the soil. Figure 1
shows the normal consolidation lines and yield sur-
faces for both unbonded and bonded materials with
various amounts of bonding.
The yield surface can be expressed as follows:

where, to obtain the behaviour shown in Figure 1,

Ps
is the preconsolidation pressure of the unbonded

material. Pc
controls the yielding of the bonded soil
in isotropic compression and p
t the cohesion and ten-
sile strength of the material; αt is a parameter. The
unbonded behaviour is recovered when b goes to Figure 1. The normal consolidation lines (NCL), critical
zero. state lines (CSL) and yield surfaces for both unbonded and
In CASM the parameters n and r control the shape bonded materials.
of the yield surface. Mθ is expressed as a function of
Lode’s angle θ, and determines the shape of the failure
surface in the deviatoric plane, following a proposal
by Sheng et al. (2002).
A classic volumetric hardening law is here
employed for the unbonded material.
Figure 2. Finite element geometry.

which has a facility to implement user-defined (UD)

∗ ∗ soil models. Numerical simulations of the expansion
λ and κ are the compression parameters of the
of an infinite cylindrical cavity were then performed.
reconstituted clay, but referred to εv :ln p
space.
Axisymmetric geometry and plane strain condi-
Bonding (b) decreases exponentially with a plastic
tions are considered. A finite outer radius, equal to
strain damage measure (h):
30 times the initial cavity radius (a0 = 1), is employed
(Figure 2). This length is enough to model the con-
dition of infinite medium for clay materials (Zentar
et al., 1998).
A strain controlled loading procedure is applied
to simulate the test load. All boundaries are sealed
h1 and h2 are material parameters (greater than zero) for drainage and no pore pressure dissipation is
defining the degradation rate. allowed.
The element used is the triangular 15-noded ele-
ment. To obtain mesh independent results more than
2.2 Numerical model
110 elements are used. In order to account for large-
The bonded soil model described above was imple- strains within the plastic zone, the analysis uses the
mented in the finite element code Plaxis Version 8, large deformations option included in Plaxis.

258
3 PROBLEM SET-UP

3.1 Bothkennar clay

The estuarine clay at Bothkennar in Scotland has
enjoyed an unusual amount of geotechnical attention
since it was selected as the UK soft clay research site
in the late 80’s. A large and thorough site investigation
and characterisation programme was performed and
its findings were firstly summarised by Hight et al
(1992) and updated with further work by Hight et al.
(2003).
The clay deposit at Bothkennar has about 15 m
thickness and is relatively uniform in terms of grading
and composition (except for organic content). Its well
known geological history justifies very little mechan-
ical overconsolidation, clearly smaller, at least, than
that apparent in oedometer tests. This apparent precon-
solidation is a first signature of bonding or structure
effects.
Another important structure signature is the differ-
ent response, while subject to the same type of triaxial
test, of samples with differing degrees of alteration.
The most carefully sampled reach higher strength and
stiffness, systematically.
Bonding in the Bothkennar clay also manifest
itself clearly in the void ratio distribution with depth
which is, despite high variability, clearly different from
that deduced from normally consolidated reconsti-
tuted samples. In other words, the in situ void index
lies generally above the intrinsic compression line
(Figure 3).
The in situ void index profile in Figure 3 shows
some abrupt changes. These are related to the differ- Figure 3. In situ void index at Bothkennar (Hight et al.,
ent units distinguished within the Bothkennar clay on 1992).
the basics of depositional and postdepositional fabric
features. Table 1. Characteristics of SBPM tests employed in this
Laboratory tests on Bothkennar clay have been study.
already modeled using generalized elasto-plastic
Depth Laboratory
bonded soil models. Baudet & Stallebrass (2004) sim- TEST (m) Holding phase data employed
ulated a few drained and undrained triaxial tests from
samples taken at two different depths. Sukolrat (2007) S3T1 2.7 No None
simulated some elaborated triaxial paths that mimic S3T2 3.1 Yes None
tube sampling effects. Both authors employed models S3T3 5.1 No OED, TXU
very similar to the one here employed albeit more gen- S3T4 6.1 Yes OED, TXU
eral in that they include anisotropic kinematic harden- S3T5 7.1 No TXU
ing, a feature not present in this model. However, the S3T6 9.1 Yes OED
hierarchical and modular nature of their elasto-plastic S3T7 10.1 No None
S3T8 12.1 Yes OED
models means that many parameters employed by
them have clear counterparts here. Hence, as explained
below, results obtained in these previous works have
been taken into account at the calibration stage. these SBPM tests took place was located very close
to the characterization study area where samples for
laboratory test have been obtained (Hight et al., 2003).
3.2 SBPM data
These tests were part of an investigation mostly
A total of eight self boring pressuremeter tests were focused on permeability measurements (Ratnam et al.,
performed at the Bothkennar site with a conventional 2002).Table 1 summarizes the test depths and indicates
SBP test of the Cambridge design. The borehole where if a dissipation strain-holding phase was included in

259
Table 2. Common parameters for all the SBPM simulations. Table 3. Specific values for initialization of each simula-
tion.
N∗ λ∗ κ∗ M n r ν h1 /h2
SBPM Depth σho uwo Ps0
2.80 0.1 0.005 1.7 3 3 0.2 1 Test (m) (kPa) (kPa) K0 (kPa) b0 h1

* N = Specific volume on the NCL for p
= 1 kPa. S3T1 2.7 53 20 0.94 12 2.0 1
S3T2 3.1 51 30 0.75 12 3.0 2
S3T3 5.1 80 50 0.60 9 5.0 3
the test. Note that the modelling approach employed S3T4 6.1 94 61 0.55 9 6.0 3.5
here did not allow us to simulate the dissipation phases. S3T5 7.1 102 75 0.43 11 5.5 3
S3T6 9.1 145 100 0.61 14 4.5 2.5
The table above includes a column for laboratory S3T7 10.1 160 106 0.67 18 3.5 2
data. It is indicated there if oedometric compressibil- S3T8 12.1 180 130 0.60 17 6.0 4
ity tests were available from a nearby sample. How
this independent data source was employed in the
simulations is explained in the following sections. the pressuremeter curve at membrane lift-off. Vertical
effective stress was then deduced from a Ko profile
3.3 Model calibration reported by Hight et al. (1992).
Hight et al. (1992) do also report a void ratio profile
It is important, at least in principle, to distinguish with depth, with notable dispersion in the measured
between model parameters and model state variables. values (Figure 3). At each test depth an average value
State variables need initialization while parameters was selected and employed to initialize the variable
need calibration. The model here employs as state- Ps
using the remoulded soil ICL. Mechanical over-
defining variables effective stress, a measure of bond- consolidation was assumed negligible for the whole
ing (b or Pc
) and a measure of density (Ps
). Model depth.
parameters will ideally represent what does not change At depths where oedometric data on intact sam-
with soil state, properties based on permanent fixtures ples was available, initial bonding (b0 ) was established
of soil (at least for the class of loads here discussed) with reference to the yield point observed on oedomet-
such as grain distribution and composition. ric conditions. At depths where such data was lacking
A set of model parameters common to all the simu- some extra assumption was needed. Here two differ-
lated SBPM tests is reported in Table 2. Parameters ent pieces of information were employed: on the one
controlling the compressibility of reconstituted soil hand the trend marked by the difference between in
(N∗ , λ∗ , κ∗ ) and the shape of the yield surface and refer- situ void index and the ICL (see Figure 3), on the
ence surface (M, n, r) were initially obtained from tests other the trend of oedometric yield stress with depth
on reconstituted clay reported by Allman & Atkinson reported by Hight et al. (1992). The resulting profile
(1992). However, it was later observed that a higher of b0 values employed in the simulations is illustrated
value of M (1.7 instead of 1.38) and a lower value of in Figure 4. The same figure reproduces the litological
κ∗ (0.005 instead of 0.01) offered better adjustment profile of the Bothkennar deposit described by Hight
to simulations of laboratory tests on natural samples et al. (1992). It will appear that the bonding profile
(as reported by Nash et al., 1992; Hight et al., 1992). inverts its tendency in unit L3, whereas the trends in
The improved parameter values thus adopted cannot be units L2 and L4 appear very similar.
justified on the basis of reconstituted behaviour only. The parameter controlling bond degradation rate,
Nevertheless, it is interesting to note that the improved h1 , was obtained comparing the oedometric curves
values of M and κ∗ are very close to those previously of reconstituted and natural material, where the latter
employed by both Baudet & Stallebrass (2004) and were available. An unexpected result was obtained in
Sukolrat (2007). that the h1 parameter that best suited the experiments
The values of Poisson ratio and, more impor- was variable with depth, varying almost proportionally
tant, the ratio h1 /h2 (controlling the relative influ- to the initial bonding value, b0 .
ence of shear and compression on destructuration) The h1 value has a relatively minor effect on the
also followed precedent (Baudet & Stallebrass, 2004; pressuremeter response, (Gonzalez, 2007) and a single
Sukolrat, 2007), by lack of any specific experimental value might have been selected for all the simulations.
information on which to sustain a different choice. However, the rule obtained from the oedometric sim-
ulation was respected and, as shown in Table 3, h1
values were selected almost proportionally to b0 even
3.4 Model initialization
for cases where no oedometric information was avail-
Table 3 reports the values initializing the state variables able. It is clear, however, that this variation points
for each SBPM test simulation. At each depth initial to a possible shortcoming of the bond evolution law
horizontal total stress and pore pressure were read from formulated in (4).

260
 250

200

Cavity pressure (kPa)

150

100

Experimental
50
Bonded clay
Unbonded clay
0
-1 1 3 5 7 9 11
-50
Cavity strain (%)
160

140

Pore pressure (kPa)

120

100

80

60
Experimental
40
Bonded clay
20 Unbonded clay

0
-1 1 3 5 7 9 11
-20
Cavity strain (%)

Figure 5. Simulation results for test S3T3 at 5.1 m depth.

Cavity strain vs cavity pressure (above) and cavity pore
pressure (below).

400

350
Cavity pressure (kPa)

300
Figure 4. Profile of initial bonding values for simulation, b0 .
Triangles indicate values obtained from oedometers. Inset: 250

litological units at Bothkennar according to Hight et al. 200

(1992). 150

100 Experimental
Bonded clay
50 Unbonded clay
4 SIMULATION RESULTS
0
-1 1 3 5 7 9 11
4.1 Pressuremeter curves -50
Cavity strain (%)
Figure 5 shows the loading and pore pressure curves 350
measured at test S3T3, performed at 5.1 depth. Along-
300
side the experimental curves, the curves labelled
Pore pressure (kPa)

bonded clay and unbonded clay represent, respectively, 250

the simulation obtained with the full model and that 200
obtained when initial bonding is artificially set to 0. 150
The curves for the bonded model show good agree-
ment on loading and unloading. The unbonded clay 100
Experimental
model results are more off the mark. 50 Bonded clay
Unbonded clay
The example shown in Figure 5 corresponds to a 0
depth where many laboratory results employed on the -1 1 3 5 7 9 11
model calibration were available (Table 1). A different -50
Cavity strain (%)
case is that of test S3T7, made at 10.1 m depth, from
which no laboratory tests were available for calibra- Figure 6. Simulation results for test S3T7 at 10.1 m depth.
tion. The results obtained on the simulation are shown Cavity strain vs cavity pressure (above) and cavity pore
in Figure 6. pressure (below).

261
 50 Cu: kPa
45 0 10 20 30 40 50 60
40 0
35
1
q (kPa)

30
25
20 2 model_bonded
15
model_unbonded
10 3
5 field_data_SBP
0 4
0 10 20 30 40 50 60 70
p' (kPa)
5

Figure 7. Stress path followed during undrained pres-

6

z: m
suremeter test by an element close to the cavity.
7
Again, the experimental cavity expansion curve is
reasonably well matched, both on loading and unload- 8
ing. However, the experimental cavity pore pressure
curve is clearly above that of the simulations. Even for 9
this more lightly bonded case, bond neglection would
lead to worst predictions. 10
A large part of the discrepancy on the pore pres-
sure curve might be attributed to the elastic part 11
of the model. Figure 7 illustrates the p
– q trajec-
12
tory of the cavity element in one simulation with the
bonded model. The introduction of bonding allows
13
a larger elastic response. The isotropic elastic model
here employed results in no pore pressure generation
Figure 8. Interpreted undrained shear strength from the
during that phase. Improvements might be obtained by field curves and the simulations.
introducing anisotropy on the elastic formulation or by
the more nuanced approach to stiffness evolution that
kinematic hardening provides. relatively simple, and certain aspects of the simula-
tion exercise have pointed to some of its shortcomings
and, more importantly, have suggested avenues for its
4.2 Interpreted results perfection.
A more compact view of the simulation results might Despite this, the results shown here prove that
be obtained using some standard approach to pres- introducing a measure of bonding, as obtained from
suremeter result evaluation. For instance, Figure 8 laboratory tests, elasto-plastic simulations of real self-
presents the result of applying the classical Gibson boring pressuremeter tests are improved. This implies
& Anderson (1961) approach to the experimental and that the self-boring procedure does not disturb the
simulated curves. original soil structure, or, if it does so, it is not in
The undrained strength values that are obtained a higher proportion to sampling procedures of the
from the bonded soil model are closer than those highest quality. Bothkennar clay is exceptional in
without bonding to the values obtained from the exper- that a large amount of high quality laboratory tests
imental curves. Clearly, this suggests that if Gibson were available. However, the precedent result supports
& Anderson undrained strength values are extracted the possibility of extracting bond-related information
from SBPM tests in a bonded soil, they will likely be from SBPM tests, something that might likely improve
higher than undrained strength values obtained from geotechnical characterisation efficiency.
tests where bonding has been damaged, for instance,
by less than perfect sampling.
REFERENCES
Allman, M.A. & Atkinson, J.H. (1992) Mechanical properties
5 CONCLUSIONS of reconstituted Bothkennar soil, Géotechnique 42, No. 2,
289–301.
Simulating test results from a whole profile of soil is Baudet, B. & Stallebrass, S. (2004) A constitutive model for
challenging. The bonded soil model employed here is structured clays, Géotechnique 54, 4, 269–278.

262
Cotecchia, F. & Chandler, R.J. (2000) A general framework Nash, D.F.T., Sills, G.C. & Davison, L.R. (1992) One-
for the mechanical behaviour of clays. Géotechnique 50, dimensional consolidation testing of soft clay from Both-
No. 4, 431–447. kennar, Géotechnique 42, No. 2, 241–256.
Gens, A. & Nova, R. (1993) Conceptual bases for a constitu- Ratnam, S., Soga, K., Mair, R.J. & Whittle, R.W. (2002)
tive model for bonded soils and weak rocks In Geotechni- Self-boring pressuremeter permeability measurements in
cal Engineering of Hard Soils-Soft Rocks, Athens, Greece, Bothkennar clay, Géotechnique 52, No. 1, 55–60.
Anagnostopoulos et al. (eds.) Rotterdam: Balkema, 577– Rouainia, M. & Muir Wood, D. (2000) A kinematic hardening
583. constitutive model for natural clays with loss of structure.
Gibson, R.E. & Anderson, W.F. (1961) In situ measurement of Géotechnique 50, No. 2, 153–164.
soil properties with the pressuremeter. Civil Engng Public Sukolrat, J. (2007) Structure and destructuration of Bothken-
Works Rev. 56, 615–618. nar clay, PhD Dept. of Civil Engineering, University of
Gonzalez, N., Arroyo, M. & Gens, A. (2007) The effect of Bristol.
structure in pressuremeter tests in clay, Numerical Meth- Yu, H.S. (1998) CASM: A unified state parameter model for
ods in Geomechanics, NUMOG X, Pande & Pietrusckzak clay and sand. International Journal of Numerical and
(eds) 721–732. Analytical Methods in Geomechanics, No. 22, 621–653.
Hight, D.W., Paul, M.A., Barras, B.F., Powell, J.J.M, Zentar, R., Moulin, G. & Hicher, P.Y. (1998) Numerical
Nash, D.F.T., Smith, P.R., Jardine, R.J. & Edwards, D.H. analysis of pressuremeter test in soil. In proceedings
(2003) The characterisation of Bothkennar clay, Charac- of 4th European Conference on Numerical Mehtods in
terisation asn Engineering properties of natural soils – Geomechanics (NUMGE), Udine: 593–600.
Tan et al. (eds.) 543–599.
Hight, D.W., Bond,A.J. & Legge, J.D. (1992) Characterization
of the Bothkennar clay: an overview. Géotechnique 42,
No. 2, 303–347.

263