Professional Documents
Culture Documents
Atkinson, JH - Non-Linear Soil Stiffness in Routine Design PDF
Atkinson, JH - Non-Linear Soil Stiffness in Routine Design PDF
5, 487±508
Soil stress-strain behaviour is highly non-linear and this has Le comportement contrainte-deÂformation du sol est extreÃme-
an important in¯uence on the selection of design parameters ment non lineÂaire et ce fait in¯uence consideÂrablement le choix
for simple routine geotechnical calculations. Non-linear be- des parameÁtres de design pour les simples calculs geÂotechni-
haviour can be characterized by rigidity and degree of non- ques de routine. Le comportement non lineÂaire peut eÃtre
linearity and these can be determined from measurements of caracteÂrise par la rigidite et le degre de non lineÂariteÂ, deux
very small strain-stiffness, peak strength and failure strain. parameÁtres qui peuvent eÃtre deÂtermineÂs en mesurant une treÁs
Very small strain-stiffness can be found from measurements faible deÂformation-rigiditeÂ, la reÂsistance maximale et la deÂfor-
of shear wave velocity in situ or in laboratory tests. Peak mation aÁ la rupture. On trouve une treÁs faible deÂformation-
strength and failure strain can be measured in routine rigidite en mesurant la veÂlocite de l'onde de cisaillement in situ
laboratory tests but are strongly in¯uenced by initiation of ou dans les essais en laboratoire. La reÂsistance maximale et la
shear bands. The important non-linear stiffness parameters deÂformation aÁ la rupture peuvent eÃtre mesureÂes par des essais
for soil are related to its composition and to its current de routine en laboratoire mais sont fortement in¯uenceÂes par
state. Back-analyses of the load settlement behaviour of full- l'initiation de bandes de cisaillement. Les parameÁtres impor-
scale and model foundations demonstrate the in¯uence of tants de rigidite non lineÂaire pour le sol sont lieÂs aÁ sa composi-
non-linear soil behaviour. Variations of stiffness with settle- tion et aÁ son eÂtat actuel. Des reÂtro-analyses du comportement
ment calculated from full-scale and model foundations agree de tassement pour des fondations grandeur nature et des
well with non-linear soil stiffnesses based on rigidity and maquettes montrent l'in¯uence du comportement non lineÂaire
degree of non-linearity. These results suggest a simple meth- d'un sol. Les variations de rigidite en fonction du tassement,
od for routine design which takes account of soil non-linear calculeÂes d'apreÁs des fondations reÂelles et des maquettes,
stiffness. correspondent bien aÁ la rigidite non lineÂaire du sol baseÂe sur
la raideur et le degre de non lineÂariteÂ. Ces reÂsultats suggeÁrent
KEYWORDS: Design footings/foundations; in situ testing; labora- une meÂthode simple pour les eÂtudes de routine, meÂthode qui
tory tests; settlement stiffness tient compte de la rigidite non lineÂaire du sol.
INTRODUCTION strengths depending mainly on drainage and strain and the peak
In the Autumn of 1969 when I started research at Imperial strength is appropriate for characterizing non-linearity. Peak
College on soil stiffness I had three textbooks. These were: Soil strengths are associated with slip planes or shear bands and it is
mechanics in engineering practice (Terzaghi & Peck, 1948), necessary to consider the in¯uence of these on strength meas-
The measurement of soil properties in the triaxial test (Bishop ured in laboratory tests.
& Henkel, 1957) and Critical state soil mechanics (Scho®eld &
Wroth, 1968). As a young research student it was dif®cult to
understand that these three books were all dealing with soils in Non-linear behaviour of soil
ground engineering. Two important themes of my work have One of the major problems in ground engineering in the
been to try to clarify the principal issues covered in these three 1970s and earlier was the apparent difference between the
books and to research soil strength and stiffness. stiffness of soils measured in laboratory tests and those back-
It is now well known that the stress±strain behaviour of soil calculated from observations of ground movements (e.g. Cole &
is highly non-linear and soil stiffness may decay with strain by Burland, 1972; St John, 1975; Wroth, 1975; Burland, 1979).
orders of magnitude. This means that for a geotechnical struc- These differences have now largely been reconciled through the
ture such as a foundation, retaining wall or tunnel, soil stiffness understanding of the principal features of soil stiffness and, in
varies both with position and with loading. particular, the very important in¯uence of non-linearity. This is
Many aspects of non-linear soil stiffness are now well under- one of the major achievements of geotechnical engineering
stood. They have been incorporated into numerical models and research over the past 30 years.
have been used with success in geotechnical design. Many of Figure 1 illustrates a typical stiffness-strain curve for soil. At
these non-linear models and numerical analyses are relatively small strains the stiffness is relatively large; at strains close to
complex and require special testing and lengthy calculation. failure the stiffness is small: this is soil being non-linear. Fig. 1
There are, however, many practical cases for which these includes typical ranges of strain for laboratory testing and for
complex models and analyses are not justi®ed and familiar structures. The ranges of strain for the different testing techni-
methods based on load factors or simple elastic analyses are ques in Fig. 1 are similar to those given by Atkinson & Sallfors
suf®cient. These may be improved if allowance is made for soil (1991). These will be discussed later in more detail. The typical
non-linearity. strain ranges for structures are those given by Mair (1993). A
The principal purposes of the 40th Rankine Lecture, and of typical characteristic strain in the ground is 0´1%; this repre-
this paper, are to consider how soil non-linearity can be sents a movement of 10 mm across a gauge length of 10 m.
quanti®ed from the results of relatively simple tests and to Generally, strains in the ground will vary from zero far away
examine the in¯uences of soil non-linearity on simple routine from the structure to relatively large values near the structure
design methods. In characterizing non-linearity it is necessary and at the edge of a rigid foundation they will be very large.
to consider both stiffness, strength and strain at failure and the The typical strain ranges proposed by Mair (1993) were based
relationships between them. Measurement of soil stiffness over on stiffnesses which gave reasonable designs for structures in
the full range of loading from very small strain to failure London Clay.
requires the use of local strain gauges but stiffness at very small
strain can be determined relatively easily from measurements of
shear wave velocity in laboratory tests or in situ. Soil has many Routine design
In geotechnical engineering there are some works which
Professor of Soil Mechanics, City University, London. require detailed analysis either because there are special design
487
Foundations
Tunnels
Local gauges
Fig. 1. Characteristic stiffness±strain behaviour of soil with typical Fig. 2. Methods for routine design of simple foundations
strain ranges for laboratory tests and structures (after Atkinson &
Sallfors, 1991 and Mair 1993)
where Är is the change of settlement due to a change of
bearing pressure, Äó , B is the width of the foundation, í is
requirements or because there are substantial economies to be Poisson's ratio, I r is an in¯uence factor which depends princi-
made. An example would be the design of a large retaining wall pally on the geometry of the foundation (Poulos & Davis, 1974)
in an urban environment. In this case, it would probably be and Es is the secant Young's modulus corresponding to the
necessary to calculate the distribution of horizontal and vertical increment of loading. Es may be related to Young's modulus for
ground movements in front of and behind the wall, stresses in very small strain Eo through a stiffness ratio Es =Eo . Again,
the wall and loads in anchors or props both during construction partial factors may be applied to account for uncertainties.
and in service. These simple routine methods may also be applied to the design
Detailed analysis and design of a major geotechnical struc- of deep foundations, retaining walls and tunnels.
ture will require special laboratory testing involving the applica- Load factor Lf and stiffness ratio Es =Eo are design para-
tion of complex stress paths and the measurement of small meters. They will depend on, among other things, the soil, its
strains together with numerical analyses using soil models state and its stress±strain behaviour, the structure and the design
which take account of the important features of soil behaviour, movements. Since these simple routine design methods aim to
including current state, recent history, in-elastic deformations, determine only one movement it must always be possible to
anisotropy, general stress states, rotation of axes of stress and select load factors or stiffness ratios which give correct solu-
strain, and so on (Hight & Higgins, 1995). All this is very tions. If the soil is non-linear then these will vary with loading
complicated and demanding and requires special equipment and and movement or strain.
expertise to obtain reliable solutions. It must be emphasized that these simple methods are, of
There are, however, very many cases where it is not so course, limited. They can work only for calculating one move-
important to have such detailed analyses and where relatively ment in one direction for relatively simple structures and well-
simple solutions are all that are needed. These routine analyses behaved soils. If more information is required, such as both
calculate only one movement in one direction; examples would vertical and horizontal movements or a pro®le of settlement or
be the settlement of a foundation, the horizontal movement at distributions of stress in the ground, then much more compli-
the top of a simple retaining wall, the surface settlement above cated analyses will be required.
the centre-line of a tunnel and so on. The simple methods described here are applicable to drained
Figure 2 illustrates the settlement of a loaded shallow or to undrained loading, making use of data from drained or
foundation and the two principal methods for routine design. undrained tests but not to cyclic or repeated loadings. They are
The general principles apply also to the design and analysis of applicable to soils which have the characteristic behaviour
simple retaining walls and tunnels. For the shallow foundation described later. These include relatively stiff ®ne and coarse
illustrated the basic requirement is to determine the design grained soils which are not strongly bonded; they exclude very
bearing pressure ó d which will cause a design settlement rd . soft soils, strongly bonded soils, and soft rocks and soils with
In the ®rst method the allowable bearing pressure ó a is unstable structure.
calculated from
1
ó a Lf ó c ó c (1) CHARACTERISTICS OF NON-LINEARITY
Fs Figure 3 de®nes basic strength and stiffness parameters for a
where ó c is a calculated ultimate bearing capacity, Lf is a load triaxial test. In Fig. 3(a) the cylindrical sample has axial and
factor and Fs is a factor of safety where Fs 1=Lf . In this radial stresses ó a and ó r and strains åa and år and the deviator
method the factor of safety or load factor is there to limit stress is q ( ó a ÿ ó r ). On loading there is a non-linear stress-
settlements; the intention is to reduce the ultimate bearing strain curve as illustrated in Fig. 3(b). The sample fails at the
capacity by a factor so that the design point is in the part of the peak deviator stress qf at a strain åf . The stiffness is Young's
load settlement curve where settlements are relatively small. modulus E which may be de®ned as a tangent Et or as a secant
Additional partial factors may be applied to various actions and Es . The stiffness at very small strains near the start of loading
reactions. is Eo .
In the second method the settlement is calculated from The tangent and secant Young's moduli vary with strain as
illustrated in Fig. 3(c). There are three regions de®ned by
Atkinson & Sallfors (1991). In the very small strain region the
Är (1 ÿ í 2 ) stiffness is approximately constant and Et Es Eo and this
Äó Ir (2)
B Es region is limited by a strain åo . There is a small strain region
qf
qf
E Eo
εf ε εr εf ε
E E
Eo
Eo
εf ε εf
εr εf = 2εr ε
(a) (b)
Fig. 5. Measurement of soil stiffness in laboratory tests Fig. 7. LVDTs used as local gauges
Bender elements
in triaxial or Down
oedometer hole
samples
Cross
hole
(a) (b)
105
Go/Gonc
Go: kPa
n
1
104
10 102 103 1 2 5 10
p ′: kPa Ro
(a) (b)
Fig. 14. Variation of Go with state for reconstituted kaolin clay: (a) normally consolidated samples; (b) overconsolidated
samples (Viggiani & Atkinson, 1995)
106
Go/Gonc
Go: kPa
105 Compacted
Truly overconsolidated
Fig. 15. Variation of Go with state for carbonate sand (Jovi i & Coop, 1997)
The carbonate sand had relatively weak grains and reached and ®ne grained soils. They also found that the values of the
states on a well-de®ned linear normal compression line at material parameters for coarse grained soils could be closely
stresses in excess of about 100 kPa above which considerable approximated by A 4000 and n 0:58, while the value of m
changes of grading were observed. Consequently, the gradings depended on the history of overconsolidation or compaction.
of truly overconsolidated samples differed from the gradings of It should be noted that, in order to determine values of
compacted samples which accounts for the different values for overconsolidation ratio Ro it is necessary to establish a true
the parameter m. normal compression line. For most coarse grained soils this will
require compression to very large effective stress (Coop & Lee,
1993).
Material parameters for very small strain stiffness
The parameters A, n and m in equation (16) are material
parameters and so they should depend on the nature of the STIFFNESS OF SOIL AT VERY SMALL STRAIN: SUMMARY
grains. Viggiani (1992) carried out bender element tests on The stiffness of soil at very small strains can be determined
reconstituted samples of a variety of different soils and her relatively simply and reliably from measurements of shear wave
results are given in Fig. 16. This shows the variations of the velocity in laboratory samples or in situ. The value of Go for a
parameters A, n and m with plasticity for ®ne grained soils. particular soil varies with current state in a simple and consis-
Although there is some scatter of the data there are clear trends tent manner given by equation (16) in which A, n and m are
showing that A decreases and both n and m increase with material parameters. For soils which are not strongly bonded or
increasing plasticity index. highly structured, these parameters depend principally on the
Coop and JovicÏic (1999) reported the results of bender nature of the grains and vary consistently with plasticity index.
element tests on a variety of different coarse grained soils. They If the soil is assumed to be isotropic, the very small strain
found that the relationships between very small strain-stiffness Young's modulus Eo can be obtained from the shear modulus
Go and state given by equation (16) applied equally to coarse with an assumed value for Poisson's ratio. If the soil is cross-
Shear
stress
0·7 Peak
Ultimate = critical state
Residual
0·5
(b)
Distortion
0·4
0·3
Coefficient, m
0·2
0·1
0 10 20 30 40 50 60 70
Plasticity index
(c)
Fig. 20. Stress characteristics for undrained loading: (a) directions of stress characteristics; (b) Mohr's circle for total
stress; (c) Mohr's circle for effective stress
2β β
Pole
δε3 δε3
β β
δε3 δε1 δε
δε1 (b)
(a) ½δγ
β ν
ν
δε3 δε1 δε
½δεv
(c)
Fig. 22. Strains in shear bands: (a) directions of zero extension; (b) Mohr's circle for constant volume straining;
(c) Mohr's circle for straining with dilation
Fig. 23. Dilation in shear bands in dry sand behind a model wall
(after Roscoe, 1970)
Distance
from soft Slip
zone in zone
observed in nominally undrained triaxial tests were all consis- inches
tent with local drainage and dilation in the shear bands.
Figure 25 shows the interpretation proposed by Atkinson &
Richardson (1987). Figs 25(a) and (c) show state paths with
axes q9, p9 and water content for overconsolidated samples with
and without local drainage. Fig. 25(b) shows the corresponding 2
stress±strain behaviour. The path ABF is fully undrained; there
is no volume change, the sample fails at F and there is no peak
strength. For the path CDE local drainage starts at D; ultimate
failure is at E, there is a peak strength at D which is the start
of local dilation. (Notice that the path ABF does have a peak
stress ratio and a peak ö9 but not a peak deviator stress.)
The peak state at B corresponds to initiation of a shear band 4
owing to local drainage in a slow undrained test. In a fast
undrained test, in which there was no local drainage and no
shear band, the peak strength corresponds to the ultimate Fig. 24. Water content observed in a shear band in the ground
strength at F. For intermediate rates of loading, or times to behind a failing wall (after Henkel, 1956)
25
20
15
0
Variation of rigidity with plasticity and state
(a) The stiffness of soil at very small strain is related to its state
1·5 by equation (16) in which the parameters A, n and m depend
on the plasticity index as shown in Fig. 16. For undrained
loading of isotropic elastic soil Euo 3Go and equation (16)
becomes
1·0 n
Euo p9
3A Rom (17)
q ′/p ′
pa pa
0·5
0 Eo/q f
0 5 10 15 20 104 103
εa: % 106
(b)
150
100
Eo: kPa
102
q ′: kPa
105
50
0
0 50 100 150 104
p ′: kPa 10 102 103
(c) q f: kPa
Fig. 27. Initiation of a shear band in a sample of overconsolidated Fig. 28. Typical values of strength and stiffness of soils (after
kaolin clay Tatsuoka & Shibuya, 1992)
Fig. 30. Variation of failure strain with state: (a) ®ne grained soils; (b) coarse grained soils; (c) general features
linearity in the range 10 to 100. In Fig. 31(a) the load factor 1·2 n l = 100
Lf q=qf . The data illustrate how load factor, Lf , and secant
stiffness ratio, Es =Eo , vary with the degree of non-linearity. At n l = 50
a strain of 0´1%, indicated by the arrows, both load factor and
stiffness ratio vary by factors of 2 to 3. n l = 20
Figures 32(a) and (b) show the variations of load factor and 0·8
stiffness ratio with the degree of non-linearity and rigidity for a n l = 10
strain of 0´1%. Over a typical range of non-linear soil parameters
Lf
the stiffness ratio needed for a design strain of 0´1% varies from Eo /q f = 1000
about 0´5 to less than 0´2 and the variation of load factor is
greater. Fig. 32(c) shows how the ratio of stiffness to strength 0·4
(Es =su ) varies with non-linearity; this is a parameter often used
in simple routine design. Similar design curves can be easily
developed for other characteristic strains from a simple stress±
strain equation, such as that given in equation (20).
The data given in Figs 31 and 32 were calculated for a value 0
of åo 0:001%. For smaller values åo has little in¯uence on the (a)
load factor or stiffness ratio. For bonded soils and soft rocks, 1·2
however, åo may be considerably larger than 0´001% and then
the value of åo begins to have an in¯uence on the load factor
and stiffness ratio. For these materials, a better basis for design
may be to avoid strains in the ground that are greater than the
value of åo , especially if the material is brittle with a rapid drop 0·8
of stiffness with strain after åo .
Es/Eo
Är (1 ÿ í 2 )
Eo /q f = 1000 Äó Ir (21)
B Es
Eo /q f = 500
where Ir is an appropriate in¯uence factor (Poulos & Davis,
Lf
0·5
1974) and, for undrained loading, íu 12. Values for the un-
drained secant Young's modulus Eus , calculated from equation
(21) from the load±settlement curves for shallow circular and
strip foundations in Fig. 33(a), are shown in Fig. 33(b) plotted
against the settlement to width ratio r=B. Also shown in Fig.
0 33(b) are values for the undrained secant Young's modulus Eus
(a) for a triaxial sample calculated from the triaxial stress±strain
1·0
curve in Fig. 33(a).
From Fig. 33(b) the values of r=B for a shallow foundation
are two to three times larger than the axial strains in a triaxial
sample at the same average stiffness. These results mean that
the stiffness at a certain strain measured in a triaxial specimen
relates to the design stiffness for a foundation at values of r=B,
E s /E o
0·5
which are two to three times larger than the corresponding axial
strain in the triaxial sample. Bolton (1993) has obtained similar
results for shallow and deep foundations and for retaining walls
using plasticity analyses.
0
NON-LINEARITY IN MODEL AND FULL-SCALE FOUNDATIONS
(b)
The in¯uence of non-linearity on foundation behaviour and
2000
on the choice of design stiffness can be illustrated by relating
the non-linear load settlement behaviour of model and full-scale
foundations to the non-linear characteristics of the soil.
1000
Figure 34 shows stiffness±strain data obtained from observa-
tions of the settlements of shallow and piled raft foundations on
London Clay made by Arup Geotechnics (1991) together with
the corresponding behaviour of London Clay in a triaxial test.
For each foundation case record the equivalent undrained secant
0 Young's modulus Eus was calculated from the bearing pressure
1 10
nl
102 103 and from the observed settlement using equation (21). In Fig.
(c) 34(a) the data are plotted as Eus =Euo and in Fig. 34(b) they are
plotted as Eus =su , which is often used to choose a value for
Fig. 32. Variation of design parameters with rigidity and degree of stiffness for routine design. Values for Euo and su were estimated
non-linearity for å 0:1% from the site investigation data for each site.
Also shown in Figs 34(a) and (b) are broken lines that
represent the behaviour of London Clay in an undrained triaxial
åa . The behaviour of the foundations is plotted as bearing test. The data for these were calculated using the simple model
pressure ó and settlement to width ratio r=B. All three curves given in equation (20) with parameters for London Clay which
approach constant stress at relatively large strains or settlements were a best estimate for the mean values for the many sites
and the bearing capacity is linked to the compressive strength considered. (The procedure used was to integrate equation (20),
through an appropriate bearing capacity factor. select a value of r to give the required values of qf and obtain
100
40
Strip foundation
80 Circular foundation
30 Triaxial test
q or σ: kPa
60
Esu: MPa
20
40
10
20
0
0 0·5 1·0 1·5 2·0 2·5 10–3 10–2 10–1 1 10
ρ/B or εa: % ρ/B or εa: %
(a) (b)
Fig. 33. Finite element analyses of shallow foundations and a triaxial test for the same soil
Piled rafts
0·8
Triaxial test
0·4
0·2
0
(a)
2000
1000
500
0
10–3 10–2 10–1 1 10
ρ/B or εa: %
(b)
values of the secant Young's modulus from the calculated Fig. 35(a). London Geotechnical Centrifuge. (b) Centrifuge model
stress±strain curve.) The solid lines in Fig. 34 are the lines for foundation
the triaxial test with strains increased three times to account for
the differences between axial strains in triaxial samples and
values of r=B for foundations.
For the foundations, the stiffnesses back-calculated from the (Stallebrass & Taylor, 1997). The foundation was 60 mm dia.
®eld observations decay with increasing values of r=B in the and during the test the centrifuge acceleration was 100 g so the
same way that stiffness decays with strain in a triaxial test. experiment was modelling a foundation 6 m dia. The instru-
The values are, however, smaller than those corresponding to ments measured the pro®le of surface settlement but only the
the line for the triaxial test with strains increased by three settlement of the rigid foundation will be considered here.
times. This is thought to be due to some drainage that probably Data from a foundation loading test are shown in Fig. 36.
occurred in the ground during construction and foundation
loading. Drainage would have the effect of increasing settle-
ments and so reducing calculated stiffnesses.
The values of r=B observed for foundations on London Clay
1·0
shown in Fig. 34 are in the range of about 0´05% to 0´5%.
These are comparable to the typical strain range for foundations Triaxial test
given by Mair (1993) and shown in Fig. 1. Triaxial strains × 3
0·8
Model test
Centrifuge model foundation on kaolin clay
Although it is always valuable to be able to compare theor- 0·6
etical analyses with full-scale observations, it is often dif®cult
Esu/Eou
to obtain all the required information about the soil, the Soil parameters:
structure, its loads and settlements and the drainage conditions. 0·4 Eou/q f = 5000
Many of these uncertainties are avoided by observation of the n l = 200
behaviour of closely monitored scale models using well docu-
mented soils. Since soils are essentially frictional materials and 0·2
many of their stiffness and strength properties depend on the
current effective stress, geotechnical models should correctly
scale effective stress. Effective stress scaling can be achieved by 0
testing models in a geotechnical centrifuge (Scho®eld, 1980). 10–3 10–2 10–1 1 10
Figure 35(a) shows the London Geotechnical Centrifuge at ρ/B or εa: %
City University (Scho®eld & Taylor, 1988) and Fig. 35(b) shows
a detail of a scale-model rigid shallow foundation on over- Fig. 36. Settlement of a centrifuge model foundation on kaolin clay
consolidated kaolin clay which was loaded in the centrifuge (data from Stallebrass & Taylor, 1997)
ρ/B ≈ 3ε
Triaxial
log ε or ρ/B