Sie sind auf Seite 1von 9

See

discussions, stats, and author profiles for this publication at:


http://www.researchgate.net/publication/245411073

A logarithmic stressstrain function


for rocks and soils
ARTICLE in GOTECHNIQUE JANUARY 1996
Impact Factor: 1.67 DOI: 10.1680/geot.1996.46.1.157

CITATIONS

DOWNLOADS

VIEWS

24

31

95

2 AUTHORS:
John Boscawen Burland
Imperial College London
88 PUBLICATIONS 2,284 CITATIONS
SEE PROFILE

Alexander M. Puzrin
ETH Zurich
106 PUBLICATIONS 830 CITATIONS
SEE PROFILE

Available from: John Boscawen Burland


Retrieved on: 31 July 2015

Puzrin, A. M. & Burland, J. B. (1996). Geotechnique 46, No. 1, 157164

TECHNICAL NOTE

A logarithmic stressstrain function for rocks and soils


A . M . P U Z R I N  a n d J. B. B U R L A N D 
function should be versatile enough to t these
data, not sacricing accuracy at large strains.

KEYWORDS: constitutive relations; laboratory tests;


numerical modelling and analysis; stiffness.

The key requirements for a stressstrain function are

INTRODUCTION

(a) the lowest possible number of constants


consistent with accuracy
(b) that the constants should have physical meaning
(c) that the constants should be easy to derive.

Numerical methods of solution of many boundary


value problems in rock and soil mechanics require
an analytical simulation of the relevant stressstrain
relationship through the entire range of strains,
from zero to strains at and beyond peak strength. In
problems of settlement and displacement of structures, the contribution of zones with very small
strains to boundary displacements can be larger
than that of zones of contained failure (see, for
example, Burland (1989)). Therefore, an accurate
simulation is required for both small strain and
close-to-failure regions. A solution is usually
implemented by means of a non-linear elastic
formulation, but other approaches may be used.
This technical note is concerned not with these
procedures but with the simplest possible mathematical representation of the experimental stress
strain curve.
A vast amount of data on small strain behaviour
of soils and rocks has accumulated since the
introduction of local strain-measuring techniques
(strain gauges, electrolevels, proximity and local
deformation transducers). However, in most of the
standard commercial tests, the resolution of strain
measurements is not sufciently high to give
reliable data in the small strain region. Therefore,
the problem of stressstrain behaviour simulation
by an analytical function should be treated
differently in the following two cases

GENERAL CONDITIONS FOR NORMALIZED


STRESSSTRAIN CURVE

Most of the published analytical stressstrain


curves were generated to t normalized experimental curves from different types of tests
(triaxial, simple shear, plane strain and so on).
Hardin & Drnevich (1972) proposed the following
expressions for normalized stresses and strains
y q=qu
x =r
r qu =Emax
where q is a stress, is a strain, qu is the ultimate
(peak) stress, r is a reference strain and Emax is the
initial stiffness modulus. Then expressions for
normalized secant and tangent stiffness are
Es =Emax y=x
Et =Emax dy=dx
Grifths & Prevost (1990) formulated ve
general conditions that a normalized stressstrain
curve should satisfy (Fig. 1).
At the strain origin x = 0

(a) case 1: small strain data are not available


the function should give acceptable accuracy
over the whole strain range using a minimum
number of parameters available from the
standard test
(b) case 2: small strain data are availablethe

condition 1: y 0
condition 2: dy=dx 1
At the ultimate strain (corresponding to the
ultimate stress) x = xu
condition 3: y 1

Manuscript received 12 December 1994; revised manuscript accepted 6 June 1995.


Discussion on this technical note closes 3 June 1996; for
further details see p. ii.
 Imperial College of Science, Technology and Medicine,
London.

condition 4a: dy=dx 0


In some circumstances it has proved convenient to
generalize conditions 3 and 4 such that at a limiting
stress qL the normalized stressstrain curve has a
157

Article number = 625

158

PUZRIN AND BURLAND

1
c

various types of loading (Table 1) are used to


illustrate these additional conditions. All the tests
used local deformation measurement techniques.
The rst four tests are reproduced from Tatsuoka &
Shibuya (1991).
Figure 2 shows the normalized secant modulus
of these soils and rocks plotted against x at small
strains. It is seen that, for the soils, the secant
modulus at x = xe decreases very rapidly, so that an
innite derivative of secant modulus can be
assumed

x=1

x = xlim

Fig. 1. Normalized stressstrain curve

d(y=x)
1
dx
For rocks, the rate of decrease of the secant
modulus at x = xe is very low, so that the condition
of zero derivative at x = xe is assumed

positive slope. Replacing qu by qL and xu by the


normalized limiting strain xL , condition 3 remains
unchanged while condition 4a is modied in
accordance with Fig. 1

condition 6a:

condition 4b: dy=dx tan (1 c)=xL


In the particular case when the limiting stress is
chosen to be equal to the ultimate (peak) stress,
= 0 (c = 1) and conditions 4a and 4b are identical.
Inside the strain interval 0 , x , xL

condition 6b:

d(y=x)
0
dx

APPRAISAL OF EXISTING CURVE-FITTING


FUNCTIONS

condition 5: dy=dx > 0

Seven widely used stressstrain functions are


examined in the light of the above conditions
(Table 2). The number at the end of a function's
abbreviation indicates the number of parameters it
requires in normalized stressstrain space. H0
indicates the original hyperbolic model, H1 its

d2 y=dx2 < 0
To improve the accuracy of the prediction at small
strains, additional conditions should be imposed on
the curve at the linear elastic strain limit x = xe .
Seven tests on different soils and rocks under
1

0.8

Es/Emax

0.6

0.4

0.2

0.2

0 .4

0.6

0.8

x
Kimachi sandstone

Toyora sand

Oya tuff

London clay

Shizuoka mudstone

Pentre silt
HPF4

Fig. 2. Small strain secant modulus for rocks and soils

Table 1. Description of tests


Material tested
Kimachi
sandstone

Oya tuff

Shizuoka
mudstone

Toyora
sand

Articial
silt HPF4

London
clay

Pentre
silt

Uniaxial
compression

Uniaxial
compression

Uniaxial
compression

Drained plane
strain
compression

Undrained HC
rotation of 19, 3 9
directions

Undrained triaxial
compression

Undrained RC + HC
simple shear

Axial stress

Axial stress

Axial stress

Stress deviator

Shear stress in
horizontal plane

Difference
between current
and initial
stress deviators

Shear stress in
horizontal plane

Axial strain

Axial strain

Axial strain

Strain deviator

Shear strain

Axial strain

Shear strain

Young's modulus

Young's modulus

Young's modulus

Shear modulus

Shear modulus

Undrained
Young's modulus

Shear modulus

qu : kPa
Emax : MPa

27 600

1440

140

6580

5216

707

xe

025

02

01

xu

191

15

19

Reference

Noma & Ishii


(1986)

Noma, Waku,
Kadota &
Murakami
(1987)

Kim, Ochi &


Tatsuoka (1990)

168
672
0008
535
Tatsuoka &
Shibuya (1991)

78

372

87

1476

620

7525

0006
84
Zdravkovic (1994)

0016
39
Standing (1994)

00003

LOGARITHMIC STRESSSTRAIN FUNCTION

Testing
conditions

47
Porovic (1994)

 HC hollow cylinder; RC resonant column.

159

160

PUZRIN AND BURLAND

modication with a correction coefcient for


strength, and H2 with correction coefcients for
both strength and initial stiffness (Kondner, 1963).
HD2, GP1 and MH5 are modied hyperbolic
models proposed by Hardin & Drnevich (1972),
Grifths & Prevost (1990) and Tatsuoka & Shibuya
(1991) respectively. J4 is the model proposed by
Jardine, Potts, Fourie & Burland (1986). For ease
of analysis, GP1 and J4 have been transformed
into the conventional normalized form. It is seen
that none of the curves satises all the conditions
specied above for soils and rocks.

A NEW STRESSSTRAIN CURVE FOR CASE 1 (L1)

As no data for the small strain region are


available in case 1, the linear elastic region has to
be neglected (xe = 0). The value of Emax (for
normalization) may be derived from dynamic tests.
An alternative approach is to estimate Emax using
existing knowledge about the inuence of various
factors (mean consolidation stress, porosity, OCR,
and so on) on initial stiffness (e.g. Tatsuoka &
Shibuya, 1991; Jardine, 1994).
The following normalized relationship (L1) is
proposed
y x x[ln (1 x)]R

(1)

Equations for the normalized secant and tangent


moduli are obtained from equation (1)
Es
1 [ln (1 x)]R
Emax

(2a)

Et
1 [ln (1 x)]R
Emax

Rx
[ln (1 x)]R1
(1 x)

The coefcients R and are dened from


conditions 3 and 4b
R

c(1 xL ) ln (1 xL )
xL (xL 1)

(3)

xL 1
xL [ln (1 xL )]R

(4)

The normalized limiting strain is never less than


unity, therefore both coefcients are positive.
Equation (1) requires only one parametereither
c for the chosen value of xL or xL = xu for c = 1 at
the peak strength. For xL = 1, curve L1 simulates
linear elastic behaviour. It can be shown that
Equation (1) satises conditions 15 for all
xL . 1. Equation (1) also satises conditions 6a
and 6b for values of limiting strain xL greater than
and less than the `critical' value xLc respectively.
This critical value of limiting strain xLc is found by
solving equation (3) for R = 1 (e.g. for c = 1,
xuc = 281449). Figure 3 indicates a family of curves
L1 for c = 1 (limiting stress equal to the peak
strength) and different values of xu .
For xu , xuc (typical for intact rocks, reaching
their peak strength at relatively small strains),
R . 1 and condition 6b is satised. In Fig. 4 the
results of the tests on rock samples are compared
with model L1. The accuracy of simulation is
remarkable considering that only one parameter is
involved. The fact that condition 6b is satised
results in a very slow change in stiffness, therefore
taking xe = 0 does not inuence the accuracy at
small strains. The L1 relationship also enables a
partial simulation of post-peak behaviour.
For xu . xuc (typical for soils), R , 1 and

(2b)

xu = 1

H0
H1
H2
HD2
GP1
MH5
J4
L1
L4

xu = 2
xu = xuc

Conditions

0.6

6a

6b

+
+
+
+
+
+
+
+
+

+
+
2
+
+
+
+
+
+

2
2
+
2
+
+
2
+
+

2
2
2
2
+
2
2
+
+

+
+
+
+
+
+
+
+
+

2
2
2
2
+
2
2
+
+

2
2
2
2
2
2
+
+
+

 Condition not satised at x = 0, but only at the linear


elastic strain limit x = xe .

Es/Emax

Model

xu = 1.5

0.8

Table 2. Conditions satised by various stressstrain


models

xu = 5
0.4

xu = 15
0.2

xu = 45
0
0

0.2

0.4

0.6

Fig. 3. Family of L1 curves

0.8

161

LOGARITHMIC STRESSSTRAIN FUNCTION


1

0.5
L1

0.5

Test
0
0

L3

(a)
0

20

40

60

(a)

0
0

0.4

0 .8

1.2

1
0.5

1.6

0.5

(b)
1
0
0

Test

0.5

L1
0
0

0.5

1.5

2.5

x
(b)

Fig. 5. Plane strain compression of Toyora sand: (a)


large strains; (b) small strains

x
(c)

Fig. 4. Uniaxial compression of: (a) Kimachi sandstone; (b) Oya tuff; (c) Shizuoka mudstone

condition 6a is satised. The results of the


simulation of the test on Toyora sand are shown
in Figs 5(a) and 5(b) for the entire strain
(0 , x , xu ) and small strain (0 , x , 3) ranges
respectively. The accuracy of simulation of this test
is comparable with that provided by the seven
functions already listed. The coefcients for GP1
and J4 were calculated using procedures given in
the corresponding papers, whereas the coefcients
for the rest of the models were taken from
Tatsuoka & Shibuya (1991). As a quantitative
criterion for accuracy, the following parameter
(increasing with increasing accuracy of prediction)
was chosen

1=2
n
1 X
a
(yi p yi e )2
(5)
n 1 i1
where yi p is predicted normalized stress at x = xi ,
yi e is experimental normalized stress at x = xi , and
n is the total number of points compared.
It can be seen from Fig. 6(a) that, for the entire
strain range, only MH5 provides higher accuracy
than L1. For small strains (x , 3), it can be seen

from
large
tions
from

Fig. 6(b) that the curve L1 (using a single


strain parameter) is bettered only by funcJ4 and MH5, which use parameters derived
the small strain data.

A NEW STRESSSTRAIN FUNCTION FOR CASE 2


(L4)

When small strain data are available, the strain


range beyond the linear elastic limit is treated as
two intervals (for smaller and larger strains)
separated by the strain x1 (where x1 = xe + e 2 1;
e = 2718281828 . . .). The stressstrain curve (L4)
is then represented by two functions that have
equal values and rst derivatives (i.e. continuous
normalized stress, secant and tangent moduli) at
x = x1 .
In the smaller strain region xe , x , x1
y xe x9 s x9[ln (1 x9)]Rs

(6)

where x9 = x 2 xe . This function requires three


parameters, which have to be derived from the
normalized experimental stressstrain curve at
small strains (Fig. 7): xe is a linear elastic strain
limit, y1 is a normalized stress at x1 = xe + e 2 1,
and ys is a normalized stress at an arbitrary value
x = xs inside the small strain interval. It has been
found that setting xs = 05 usually gives a good t
for small strains.

162

PUZRIN AND BURLAND

where

x x1
xL x1
g; xL 99
g
1 y1
1 y1



dy
Rs
g
1 s 1 Rs
dx
e

120

x99

80

xx1

40

Coefcients RL and L are dened from conditions


3 and 4b

0
H0

H1

H2

HD2 GP1
(a)

J4

MH5

L1

L3

RL

160

120

80

40
0
H0

H1

H2

HD2 GP1
(b)

J4

MH5

L1

L3

Fig. 6. Appraisal of the accuracy of various models for


simulating stressstrain behaviour of Toyora sand: (a)
for large strains; (b) for small strains

c
y1
y
ys
xe
0
0

[c (1 c)(x1 =xL ) y1 ]
1 y1

xe

xs

xe = xe + e 1

xlim

Fig. 7. Parameters of the L4 curve

Then coefcients s and Rs are dened


from the conditions that the function should
pass through the points (x1 , y1 ) and (xs , ys )
respectively
x1 y1
s
(7)
e1
Rs

ln (xs ys ) ln [s (xs xe )]
ln [ln (1 xs xe )]

In the larger strain region x1 , x , x1


y y1
x99 L x99[ln (1 x99)]RL
1 y1

(8)

(9)

(1 xL 99) ln (1 xL 99)
xL 99(xL 99 1)

xl 99 1
xL 99[ln (1 xL 99)]RL

(10)
(11)

Equation (9) requires one additional large strain


parameter: either c for the chosen value of xL or
xL = xu for c = 1 at the peak strength. Relationship
L4 satises conditions 15 always and either 6a or
6b depending on the value of Rs . Relationship L4
can be simplied, assuming xe = 0 and choosing a
limiting stress equal to the peak strength. This
modication is presented as a solid line in Fig. 5,
simulating the test on Toyora sand as L3 with only
three parameters. The original function L4 gives a
better prediction, but L3 was used in order to allow
a comparison with other functions, which also
neglect the elastic region in simulating this test. It
is seen from Fig. 6 that L3 ts both the large and
small strain test data with the highest accuracy of
all the models, including those with four and ve
parameters.
Relationships L1 and L4 were used to simulate
various types of tests on London clay, Pentre silt
and articial silt HPF4 as indicated in Figs 810
respectively. The accuracy of simulation is
seen to be remarkable for L4 and very acceptable
for L1 over both the small and entire strain
regions.
CONCLUSIONS

The conditions for a typical non-linear stress


strain function for soils and rocks are specied,
and existing functions are shown to be decient in
many of them. A new logarithmic function is
proposed which is shown to satisfy all the general
conditions.
The function can be tted to normalized
experimental stressstrain data using one, three
or four parameters depending on the availability
of small strain measurements. All the parameters
have physical signicance and are easy to derive.
Comparison with the results of different types of
tests on a wide range of rocks and soils shows
high accuracy of simulation over the entire strain

163

LOGARITHMIC STRESSSTRAIN FUNCTION


1

0.8

0.8

0.6

0.6

0.4

L1

0.4

L1

Test
0.2

L3

0
0

10

15
x
(a)

20

25

30

L4

0
0

x
(a)

0.4

0.5
0.4

0.3

Test

0.2

0.3

0.2

0.2
0.1

0.1
0
0.001

0.01

0.1

Log x

0
0.001

0.01

(b)

Log x

0.1

(b)

Fig. 8. Triaxial compression of London clay: (a) large


strains; (b) small strains

Fig. 10. Hollow cylinder test on articial silt (HPF4):


(a) large strains; (b) small strains

range. Even in its single parameter form the


function gives remarkably good simulation.

1
0.8
0.6

ACKNOWLEDGEMENTS

0.4

The work was supported by the UK Foreign and


Commonwealth OfceClore Foundation Scheme
through the British Council.

L1
Test

0.2

L4

0
0

10

20

30

40

x
(a)
0.4

0.3

0.2

0.1
0
0.0001

0.001

0.01
Log x
(b)

0.1

Fig. 9. Simple shear of Pentre silt: (a) large strains;


(b) small strains

NOTATION
c parameter dened in Fig. 1
Emax initial stiffness modulus
n number of points compared
q stress
qL limiting stress
qu ultimate (peak) stress
R coefcient dened in equation (3)
xe linear elastic strain limit
xL normalized limiting strain
x1c critical limiting strain
yi e experimental normalized stress
yi p predicted normalized stress
yL normalized stress at xL = xe + e 2 1
ys normalized stress at x = xs
coefcient dened in equation (4)
parameter dened in Fig. 1
strain
r reference strain

164

PUZRIN AND BURLAND

REFERENCES
Burland, J. B. (1989). Small is beautifulthe stiffness of
soil at small strains. Can. Geotech. J. 16, No. 4, 499
516.
Grifths, D. V. & Prevost, J. H. (1990). Stressstrain
curve generation from simple triaxial parameters. Int.
J. Numer. Analyt. Methods Geomech. 14, 587594.
Hardin, B. O. & Drnevich, V. P. (1972). Shear modulus
and damping in soils: design equations and curves.
J. Soil Mech. Fdn Engng Div. Am. Soc. Civ. Engrs 98,
No. 7, 667692.
Jardine, R. J. (1994). One perspective of the prefailure
deformation characteristics of some geomaterials.
Proceedings of international symposium on pre-failure
deformation characteristics of geomaterials, Sapporo.
pp. 151182. Rotterdam: Balkema.
Jardine, R. J., Potts, D. M., Fourie, A. B. & Burland, J. B.
(1986). Studies of the inuence of non-linear stress
strain characteristics in soilstructure interaction.
Geotechnique 36, No. 3, 377396.
Kim, Y.-S., Ochi, K. & Tatsuoka, F. (1990). Strength
and deformation properties of mudstones in triaxial

compression (in Japanese). Proc. 8th Jap. Symp. Rock


Mech., 357362.
Kondner, R. L. (1963). Hyperbolic stressstrain response:
cohesive soils. J. Soil Mech. Fdn Engng Div. Am.
Soc. Civ. Engrs 89, No. 1, 115143.
Noma, T. & Ishii, T. (1986). On measuring method of
axial strain in triaxial test on rock (in Japanese).
Proc. 41st Annual Conf. Jap. Soc. Civ. Engrs 3, 657
658.
Noma, T., Waku, A., Kadota, S. & Murakami, H. (1987).
On measuring method of axial strain of rock specimen by means of proximeter transducers (in Japanese). Proc. 20th Jap. Symp. Rock Mech. 6165.
Porovic, E. (1995). Investigation of soil behaviour using
a resonant column torsional shear hollow cylinder
apparatus. PhD thesis, University of London.
Standing, J. R. (1994). Private communication.
Tatsuoka, F. & Shibuya, S. (1991). Deformation characteristics of soils and rocks from eld and laboratory
tests, Proc. 9th Asian Conf. Soil Mech., Bangkok 2,
101170.
Zdravkovic, L. (1994). Private communication.

Das könnte Ihnen auch gefallen