R. Baker1
Abstract: A substantial amount of experimental evidence suggests that Mohr envelopes of many soils are not linear, particularly in the
range of small normal stresses. This departure from linearity is significant for slope stability calculations since for a wide range of
practical stability problems, critical slip surfaces are shallow and normal stresses acting on such surfaces are small. The present work
presents a procedure for estimating parameters of the Mohr form of Hoek and Brown empirical failure criterion, based on triaxial data.
This nonlinear Mohr envelope provides significantly better representation of experimental information than the Mohr–Coulomb criterion
in cases when the data set includes tests at small enough normal stress. Application of the estimation procedure to published data sets
shows that effective cohesive strength of most soils is close to zero, and Mohr envelopes are continuous at the tensile strength limit.
DOI: 10.1061/共ASCE兲1090-0241共2004兲130:5共498兲
CE Database subject headings: Envelope; Tensile strength; Data analysis; Stress.
冉 冊
one month, a written request must be filed with the ASCE Managing
Editor. The manuscript for this paper was submitted for review and pos- n
sible publication on September 26, 2002; approved on April 11, 2003. ⫽S NL共 兩 A,n,T 兲 ⬅ P a A ⫹T (1)
Pa
This paper is part of the Journal of Geotechnical and Geoenvironmental
Engineering, Vol. 130, No. 5, May 1, 2004. ©ASCE, ISSN 1090-0241/ where P a stands for atmospheric pressure; 兵 A,n,T 其
2004/5-498 –506/$18.00. ⫽nondimensional strength parameters; and S NL( 兩 A,n,T)
⫽SMC共 兩 c⫽0, 兲
⫽S NL共 兩 A⫽tan共兲,n⫽1,T⫽0 兲
2. For n⫽1/2 Eq. 共1兲 implies 2 ⫽(S NL()) 2 ⫽( P a A) 2 (/ P a
⫹T). This relation is a generalization of Griffith’s yield cri-
terion 2 ⫽4t(⫹t) 共Griffith 1924兲 and the original form of
Griffith’s law is the following limiting case of the NL crite-
rion:
S GRF共 兩 t 兲 ⬅2 冑t 共 ⫹t 兲 ⫽S NL共 兩 A⫽2 冑t/ P a ,n⫽1/2,T⫽t/ P a 兲
The above observations show that the NL strength law defined by
Eq. 共1兲 is a generalization of the classical MC and Griffith mod-
els; including them both as particular limiting cases. This property
of S NL() makes it a very useful and versatile constitutive as-
sumption.
Fig. 1. Schematic strength function ⫽S() The tangential friction angle, t associated with S NL(), is
defined by the relation
dS NL共 兩 A,n,T 兲 An
tan关 t 共 兩 A,n,T 兲兴 ⬅ ⫽ (2)
d 共 / P a ⫹T 兲 共 1⫺n 兲
⫽nonlinear 共NL兲, strength function defined by Eq. 共1兲. The ex- The geometrical significance of t is introduced in Fig. 1. Eq. 共2兲
plicit notation S NL()⫽S NL( 兩 A,n,T) emphasizes the depen- shows that t (→⫺ P a T)→/2. Consequently, the NL strength
dence of strength functions on their parameters 共quantities written model considered in the present work is continuous at the tensile
on the right of the vertical line are considered as given constants兲. strength limit, and it does not have the characteristic discontinuity
The nonlinear strength criterion for rock masses 共Hoek and of both the MC and PF models.
Brown 1980兲 is associated with a Mohr envelope that has similar
character to the one defined in Eq. 共1兲. Expressing Hoek and Transformation between Mohr and Triaxial Planes
Brown’s 共HB兲 criterion, as a Mohr envelope is convenient for
limiting equilibrium applications 共e.g., slope stability兲. An addi- The basic experimental variables obtained from triaxial tests are
tional advantage of this form is that the parameters 兵A,n,T其 have the minor and major principal stresses at failure 兵 3 , 1 其 . These
clear physical significance. In particular, A is a scale parameter experimental variables define the combinations 兵 p⫽( 1
controlling the magnitude of shear strength, T is a shift parameter ⫹ 3 )/2,q⫽( 1 ⫺ 3 )/2其 . The collection of all possible 兵p,q其 val-
controlling the location of the envelope on the axis 关 t NL ues is designated as the experimental triaxial plane.
⬅ P a T is the tensile strength associated with S NL() Consider the problem of establishing the MC strength param-
⫽S NL( 兩 A,n,T), with T representing a nondimensional tensile eters 兵 c, 其 based on a collection of m given data points 兵 p i ,q i 其
strength兴, and n controls the curvature of the envelope. i⫽1¯m. The MC criterion ⫽S MC( 兩 c,)⫽c⫹ tan() can be
In order to guarantee that S NL( 兩 A,n,T) is consistent with the written in the form
general requirements specified in ‘‘Nonlinear Mohr Envelopes– q⫽Q MC共 p 兩 a,b 兲 ⫽a⫹bp (3a)
General Considerations’’ it is necessary to impose the restrictions where
兵 A⬎0,1/2⭐n⭐1,T⭓0 其 . The inequality n⭐1 follows from the
a⫽c cos共 兲 and b⫽sin共 兲 (3b)
convexity requirement. The restriction n⭓1/2 follows from the
observation that for n⬍1/2 the radius of curvature of S NL() is Given the data set 兵 p i ,q i 其 i⫽1¯m, and the representation q
less than the radius of the tangential Mohr circle 共Jiang et al. ⫽a⫹bp, it is possible to establish the parameters 兵a,b其 by least
2003兲. Under such conditions 共i.e., n⬍1/2), S NL() intersects squares curve fitting, and evaluate the MC strength parameters as
twice with the same Mohr circle, thus contradicting the basic ⫽arcsin(b) and c⫽a/cos(). The feasibility of this process de-
definition of Mohr envelopes. The original form of the HB em- pends on availability of the explicit representation 关Eq. 共3a兲兴 of
pirical criterion was derived without using the notion of Mohr the MC criterion in the experimental 兵p,q其 plane. This technique
envelopes, and the extensive literature dealing with this criterion of establishing MC strength parameters is well known, and it is
does not include the requirement n⭓1/2. It is noted however that described here mainly in order to contrast it with the problem of
this requirement is satisfied in applications of this criterion to real establishing the strength parameters 兵A,n,T其 of the NL strength
experimental data 共e.g., Hoek and Brown 1980; Hoek 1983; law.
Sheorey 1997兲. Using Eqs. 共1兲 and 共2兲 and considering the geometry of Mohr
It is instructive to discuss the following limiting cases de- circles 共Fig. 1兲 it is possible to verify that
scribed by the nonlinear strength law introduced in Eq. 共1兲: 1 ⫹ 3
1. Consider the limiting case n⬅1. In this case S NL( 兩 A,n,T) p⬅ ⫽⫹ tan共 兲
2
is reduced to the linear MC criterion; resulting with the re-
lation S MC( 兩 c,)⬅c⫹ tan()⫽SNL关 兩 A⫽tan(),n⫽1,T ⫽⫹S NL共 兲 tan„ t 共 兲 …
⫽c/ P a tan()兴. The limiting case of a restricted (c⫽0), MC
material is frequency used in practical slope stability calcu-
lations, and the designation purely frictional 共PF兲 will be
used when referring to this model. The PF model is the fol-
⫽⫹ P a nA 2 冉
Pa
⫹T 冊 共 2n⫺1 兲
⫽S NL共 兲 冑1⫹tan2 共 t 共 兲兲
⫽ P aA 冉
Pa
⫹T 冊冑
n
1⫹ 共 An 兲 2 冉
Pa
⫹T 冊 2 共 n⫺1 兲
⬅q 共 兩 A,n,T 兲 (4b)
The following observations are relevant with respect to Eq. 共4兲:
1. Inspection of Eq. 共4兲 shows that p(⫽⫺ P a T 兩 A,n,T)
⫽⫺ P a T and q(⫽⫺ P a T 兩 A,n,T)⫽0, i.e., the tensile
strength point T has the same location in the 兵,其 and 兵p,q其
planes.
2. Inverting Eq. 共4兲 results with 3 ( 兩 A,n,T)⫽p( 兩 A,n,T)
⫺q( 兩 A,n,T) and 1 ( 兩 A,n,T)⫽p( 兩 A,n,T)
⫹q( 兩 A,n,T). Consequently, using Eq. 共4兲 it is possible to
evaluate principle stresses implied by the NL law at each
value.
3. The general relations 兵 p⫽p(),q⫽q() 其 关Eq. 共4兲兴 show
that the nonlinear strength function S NL( 兩 A,n,T) has a
parametric 共rather than explicit兲 representation in the experi-
mental 兵p,q其 plane, and it cannot be represented as a function
q⫽Q NL(p 兩 A,n,T). As a result, the simple process used for
determination of the MC strength parameters cannot be ap-
plied in the present case. This difficulty was discussed by a
number of writers 共Ucar 1986; Sheorey 1997兲. An iterative
estimation process resolving this technical difficulty is pre-
sented in the next section. However, at this stage it is instruc-
tive to compare the representations of the NL law in 兵p,q其
and 兵,其 planes.
The function q⫽Q NL(p) exists 共despite the fact that it does not
have an explicit representation兲. It is meaningful, therefore, to Fig. 2. Representations of NL strength functions in Mohr and triaxial
define the derivatives dq/dp⫽Q NL ⬘ (p) and d 2 q/dp 2 ⫽Q NL ⬙ (p) of planes 兵 A⫽1,T⫽0.1其
this function, and study their properties. Straightforward algebraic
manipulations result with the following relations:
⬘ 共 p 兩 A,n,T 兲 ⭐1
0⭐Q NL 2. All pairs 兵p,q其 located on the limiting line q⫽Q Lmt(p 兩 T) are
mapped into the tensile strength point T⬅ 兵 ⫽⫺t,⫽0 其 in
⬘ 共 p⫽⫺ P a T 兩 A,n,T 兲 ⫽1
Q NL (5a) the Mohr plane, i.e., this limiting line is a singularity of the
transformation between the Mohr and triaxial planes. In prin-
⬘ 共 p→⬀ 兩 A,n,T 兲 ⫽0
Q NL
ciple Q NL(p 兩 A,n,T)⬍Q Lmt(p 兩 T) for all p⬎⫺ P a T, however
⬙ 共 p 兩 A,n,T 兲 ⭐0
Q NL there are p ranges in which these two functions are
(5b) practically identical 关e.g., Fig. 2共b兲 shows that
⬙ 共 p⫽⫺ P a T 兩 A,n,T 兲 ⫽0
Q NL Q NL(p 兩 A,n⫽0.5,T) is almost identical to Q Lmt(p 兩 T) in the
range between points T and a兴. In such ranges the transfor-
Eqs. 共5a兲 and 共5b兲 are identities, valid for all legitimate param- mation between 兵p,q其 and 兵,其 planes become numerically
eters 兵A,n,T其 and all legitimate values of p. The first of Eq. 共5b兲 ill conditioned, requiring careful evaluations.
shows that Q NL(p 兩 A,n,T) is a convex function. Eq. 共5兲 implies
that there exists a limiting line q⫽Q Lmt(p 兩 T)⫽p⫹ P a T in the
兵p,q其 plane, and all legitimate 兵p,q其 pairs satisfy q⭐Q Lmt(p 兩 T). Curve Fitting the N – L Strength Function to Triaxial
The validity of this conclusion is illustrated in Fig. 2, which Data
shows NL strength functions for 兵 A⫽1,T⫽0.1其 plotted in Mohr
兵,其 and triaxial 兵p,q其 planes, respectively. This figure illustrates Experimental results of a triaxial testing program can be repre-
the following points: sented by the array 兵 p i ,q i 其 i⫽1¯m, while the nonlinear
1. The function ⫽S NL( 兩 A,n,T) has two characteristic points strength function S NL( 兩 A,n,T) is defined in the Mohr plane 兵,
T and B at which strength does not depend on n. The char- 其. In order to perform curve fitting it is necessary to have both
acteristics of the tensile strength point T were discussed pre- data and function in the same plane. In the present work we chose
viously. Point B is located at 兵 ⫽ P a (1⫺T),⫽ P a A 其 , and to transform the experimental information into the Mohr plane.
tan关t(⫽Pa(1⫺T))兴⫽An. ⫽ P a (1⫺T) is a convenient ref- Considering the geometry of Mohr circles 共Fig. 1兲, it is possible
erence value defining a boundary between small and large to write
normal stresses. In the range ⬍ P a (1⫺T) strength de-
creases with n while for ⬎ P a (1⫺T) it decreases with n. i ⫽p i ⫺q i sin共 i 兲 (6a)
Example Problems
冉冑 冊
ing qualitative discussion, and from now on we will refer
simply to an ‘‘experimental normal stress range’’ without 1 ⫹t
s ⬅2 arctan ⫺ (8)
specification of a model. 3 ⫹t 2
5. The MC criterion is linear and the proposed NL law is con-
vex. This has a number of practically significant implications
with respect to performance of these two laws; in particular: The following observations are relevant with respect to this rela-
• Fitting both laws to the same data set, the NL model pre- tion:
dicts slightly larger strength values than MC in the central 1. Envelopes associated with s and t are tangential to a
part of experimental normal stress range. This second given Mohr circle at different points 共points B and A in Fig.
order effect usually decreases with the number of observa- 1兲. The tangential friction angle t is the physically signifi-
tions. cant variable controlling slip surface inclination, and s is
• A second, and much more important consequence of con- used only to compare predictions of the NL model with ex-
vexity, is that the NL model predicts significantly smaller perimental information, without associating a physical sig-
strengths than MC at values outside of the experimental nificance to it.
normal stress range. The practical purpose of the NL 2. Eq. 共8兲 is formally similar to a well-known relation usually
strength function is not to provide an improved curve fit- derived in the MC framework, but it has a different physical
ting of experimental strength information 共although the fit- significance. In the MC framework t⫽t MC⫽c/tan() is a
ting is usually improved兲. Rather, the practical significance derived quantity, while in Eq. 共8兲 t is a given tensile strength.
of this law is that it implies an alternative 共more conserva- An important feature of Eq. 共8兲 is that s ⫽ s ( 3 , 1 兩 t),
tive than MC兲, extrapolation process for transferring ex- i.e., for a given value of tensile strength, tangential friction
perimental strength information to zones outside of the ex- angles defined by Eq. 共8兲 depend on the experimental vari-
perimental normal stress range, where direct experimental ables 兵 3 , 1 其 and they are not associated with any particu-
information is not available. The engineering significance lar strength function S().
of this observation was discussed by Baker 共2003兲. 3. Soils are particulate media, and tensile strengths of such ma-
terials are either zero or very small 共approaching zero兲. Con-
sidering zero tensile strength materials, Eq. 共8兲 is reduced to
Example Problem 2—Dune Sand
s ⬅2 arctan(1 /3)⫺/2. In this particular case secant fric-
The Mohr circles in Fig. 5共a兲 are results of drained triaxial tests tion angles are a measure of the principle stress ratio at fail-
on dune sand from southern Israel 共Frydman 1968兲. Sand is usu- ure; and those angles are fixed by the experimental variables
ally considered as the classical purely frictional material, and in 兵 3 , 1 其 . The points in Fig. 5共b兲 are experimental values of