Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

Panel discussion: Definition and determination of characteristic soil properties

Débat de spécialistes: Définition et détermination des valeurs caractéristiques des propriétés


des sols

H. R .S chneider - Geo-Consulting, Zug, Switzerland

A B S T R A C T : The fundamental soil parameter used in E C 7 "G eotechnics" is the characteristic value Xk, from which the design value x j is
derived. These soil parameters are defined and compared to todays practice. A simple approach, derived from fundamental statistics and
probability concepts, is proposed for the selection o f the characteristic values. The approach presented in this paper is entirely based on
today's practice. It offers an extension to incorporate experience or judgmental knowledge as well as test values to produce characteristic
values Xk. The resulting characteristic values are equally valid using the EC7-approach as well as for today's method. The proposed
equation ( 6 ) is derived from fundamental concepts o f statistics and probability, yet the design engineer does not need to acquire those
concepts to successfully employ the method presented. An illustrative example to demonstrate the application o f method is given.

1 IN T R O D U C T IO N 2 .2 E u r o c o d e 7: P a r t ia l sa fe ty f a c t o r s

Civil engineering design is largely a matter o f decision making E u rocod e 7 uses the limit state format with a set o f fixed partial
under uncertainty. The loads and responses are not known factors o f safety, which is at present considered the most rational
exactly, the accuracy o f the design method is uncertain, and the design procedure for geotechnical routine problems. Within this
choice betw een alternative designs or construction procedures is fram ework the characteristic values o f soil properties have to be
rarely clear cut. At all stages from initial concept to final selected in a well defined way if a specific safety level shall be
com pletion, decisions must be made using either incomplete obtained. Variability or uncertainty in soil layers is only taken
information or super-abundant and inconsistent information. This into account in selecting the characteristic soil values
inherent uncertainty is particularly true for geotechnical accordingly.
engineering since natural soils are extremely variable in their
properties and the rational selection o f suitable design parameters T he characteristic value Xk is therefore the
is generally one o f the most difficult part o f a design. Little fundamental soil parameter in E C 7
guidance can be found in standard soil mechanics textbooks.
F o r geotechnical design tasks some kind o f a conservatively B oth, the characteristic value Xk (E C 7) as well as the
chosen mean value, often based on local experience or subjective conservatively chosen mean value (traditional approach) are very
information, needs to be selected. I f not entirely based on expe­ closely related and are assumed to be equivalent for practical
rience only, in practice the information on the mean properties is purposes.
sometimes supplemented by relatively few test values.
In this paper, practical definitions o f the fundamental soil
param eters as used in E C 7 "G eotechnics", such as the 3. S U B S O IL M O D E L L IN G
characteristic value xk and the design value x<i, are presented. In
addition, a simple approach, derived from fundamental statistics There are tw o strongly interdependent tasks to be solved for
and probability concepts, is proposed for the selection o f the each geotechnical problem:
characteristic values.
• Selection o f layers
B y means o f this approach, the characteristic soil properties
T he subsoil is idealized into a few well defined and
can be determined with subjective experience only, based on test
hom ogeneous soil layers, which constitute the basis o f
results only, as well as with the combined information o f the two.
subsequent analyses. This is by no means a trivial step and
An illustrative example to demonstrate the application o f method
generally involves an iteration process between the two
is given.
points described here.

• Selection o f soil properties for each layer


2 S A F E T Y IN G E O T E C H N IC A L D E SIG N M E T H O D S T he “averaged“ soil properties (= characteristic values) are
selected for each layer taking into account factors such as:
2.1 T r a d itio n a l A p p r o a c h : G lo b a l sa fe ty f a c t o r s
- geological, hydrogeological and other pertinent
background information e.g. from previous projects
T he traditional or current design method is based on the global
safety concept. All variables such as loads, soil parameters, - T he variability o f the soil properties within each layer
calculation procedures are used at a level o f a cautiously selected
- T he extent o f the zone o f ground influencing the
mean for the design computations. A global factor o f safety,
behaviour o f the geotechnical structure
accounting for the combined variability and uncertainty effects, is
applied to arrive at a reasonable safety level o f the final - The specific geotechnical problem to be solved, e.g.
geotechnical structure. peak shear strength versus residual strength, etc.
T he fundamental soil properties used for today's design
computations are conservatively chosen mean values. Their In the following we will mainly focus on the selection and deter­
degree o f conservatism is generally not defined on a quantitative mination o f “averaged“ soil properties for each selected layer.
basis, but depends on the experience and judgment o f the T he term “averaged“ soil value, as used by Terzaghi (1 9 4 0 ),
designer. today is synonymous with “conservatively chosen mean vaiue“ .

2271
4. C H A R A C T E R IS T IC SO IL P R O P E R T IE S
Xk = xm - standard error o f means
In principle it would be possible to measure the soil properties
everywhere in the affected volume and determine the statistical
mean values o f that zone. In this case we would know the true
mean value o f the selected layer and could use it for further c) D es ig n V alue Xj
analyses.
In practice, however this procedure is quite impossible. In T he design values x j for soil parameters are generally derived
reality, therefore the true mean properties o f a layer do remain from the characteristic values xk
unknown and estimates o f the true mean have to be made on the
basis o f experience or only small numbers o f tests.
( 1)
T he true mean properties o f a layer do remain unknown
and estimates o f the true mean have to be made on the Ym is the partial factor o f safety as given in E C 7 (1 9 9 4 ), Table
basis o f experience or only small numbers o f tests 2 . 1.

4 .2 D eter m in a tio n o f c h a r a c te r is tic s o i l p r o p e r tie s


4 .1 D efin itio n s a n d p r a c t i c a l m ea n in g s
Based on statistical sampling and estimation theories, the
The different meanings o f the above mentioned values as well as characteristic value xk can be described by the following general
the soil parameters referred to in Eurocode 7, will be defined and expression:
also explained in practical terms:
S u m p le

• mean value xm Xk = Xm - f •--- ~F=- (2)


Vn
• characteristic value Xk
• design value x j

a) M e a n V alu e xm

T he mean value o f a soil property corresponds to the true


where: S„™pie =
standard deviation o f the test values
statistical mean o f a soil layer and is usually unkown due to
n =
number o f samples or test values Xj
limited sampling. It is however this mean value which we are
f =
statistical coefficient related to the type
targeting at, since it is a clearly defined fundamental soil
o f distribution (Normal, Stu dent's....),
parameter to be used for design computations. confidence limits and number o f test
The best we are able to do is to estimate the unknown mean
values
value o f the layer as exactly as possible with the best “limited“ Substituting the coefficient o f variation, V x = S „ mpie/xm, into
information at hand. This estimation does involve some uncer­
equation ( 2 ) gives
tainty which needs to be accounted for accordingly.
The mean value itself is not used in Eurocode 7. Instead the
characteristic value, being the best possible estimate o f this mean Xk = Xm •^1 - —
J = ■V xj H)
value o f the selected layer, is used in place o f the mean.

Although this statistical equation is correct for our problem at


b) C h a r a c te r is tic V alue Xj, hand, it poses several difficulties to apply it directly in
geotechnical practice. T he most serious limitations are shown in
The characteristic value xk is the best estimate Figure 1 below.
o f the “unknown“ statistical mean xm o f a soil layer

A ccording to E C 7 : The characteristic value shall be selected with


f
the aim that the probability o f a more adverse (mean) value
Jn
governing the behaviour o f the soil and rock in the ground is not
greater than 5% .

W hat exactly does this mean?

Characteristic value Xk = a conservatively chosen


mean value o f the layer
or
a cautiously selected mean
value o f the selected soil
layer num ber o f test values n

In E C 7 the degree o f conservatism or caution is further defined Figure 1


as follow s:
• The probability o f a more adverse real mean value governing
In the range o f n = 0 - 1 test values, statistical methods cannot be
the behaviour o f the soil layer should not be higher than 5% .
used (standard deviation is not defined). This however is a fairly
• W e want to be approximately 95 % confident that the real typical situation for smaller geotechnical projects.
statistical mean o f the layer is superior to the selected Xk. In the range o f n = 2 - 9 test values, statistical methods could

2272
be used theoretically. As demonstrated in Figure 1 however, the experience or judgment, documented local or regional experience
coefficient on the left axis exhibits a value too high due to the as well as published tables with typical values.
small and uncertain statistical basis and stabilizes only after a The value o f Xm can reliably be estimated by the above method,
larger number o f n. In addition, the coefficient o f variation V , on since it is fairly insensitive to heterogeneity or scatter around the
purely statistical grounds will also be somewhat higher than mean properties.
typically encountered in soils. Since both these coefficients are T he coefficient o f variation Vx however poses more difficulties
multiplied, as seen in equation (4 ), the resulting characteristic since it is directly dependent on the difference between the
value would be unreasonably low and pessimistic. minimum and maximum values o f a and c. The above equations
In the range o f n > 10 test values, statistical methods can be (7 ) and ( 8 ) implicitly assume that a and c are 3 standard
applied sucessfully. It is however rarely the case to have such a deviations low er or higher, respectively than the mean value x,„.
large data base in geotechnical projects. This means that approximately 100% o f the possible data should
Com parative computations have shown that a good be enclosed by the range o f a to c. Measured values do usually
approximation to the above formula is achieved with not represent 1 0 0 % o f the statistically possible data, but often
only about +/- 2 standard deviations, which should be accounted
for.
T able 1 presents typical values for the coefficient o f variation
V x. They are generally valid and their magnitudes have been
confirmed by many researchers worldwide, (e.g. Rethati, 1988;
This approximation does produce characteristic values within Lum b, 1974).
reasonable accuracy and is valid for several distributions typical
for soils. It also compares very favourably with values typically
used in the geotechnical practice. A m ajor advantage is that the It has been found on a worldwide data basis that
method can also be used when no test values at all are available V„ is approximately constant for a certain soil property
as is often the case.

Proposed relationship for the determination o f xt ______^ Therefore we always have very valuable prior
information available for estimating the characteristic
soil properties, even in the absence o f any test values.
Vx
Xk = Xm •I 1 - — (6 ) Such a priori values are presented in Table 1.

T he equation means, that half o f a standard deviation needs to be


Table 1
deducted from the mean value Xm to obtain the characteristic
value Xk. As will be seen in chapter 4 .2 .a), it is advantageous to
Soil Property Range of Recommended
w rite the equation in terms o f V x and not with Sx.
typical values of Vx average values V„
This equation has been proposed in Switzerland 6 years ago
(Schneider 1990). Its use in practical applications has proven its
Densities 0 .0 1 - 0.1 0 0
simplicity and m ore important, close correlation with values
estimated by experienced geotechnical engineers. Angle o f internal Friction 0 .0 5 -0 .1 5 0 .1

Equation ( 6 ) can easily be applied e.g. to scattered data as a


Cohesion 0.3 -0 .5 0.4
function o f depth, since half o f a standard deviation can either be
determined visually or by means o f statistical regression analyses. Compressibility modulus 0 .2 - 0 .7 0.4

4.3 D eter m in a tio n o f th e v a lu es xm a n d Vx


b) T est v a lu es a v a ila b le w ithou t a p r io r i in form ation

As seen by inspection o f the proposed equation ( 6 ), only the two


In this case, the 2 required values xm and V„ are calculated using
soil parameters xm and V x are required for the assessment o f the
standard equations from statistics, as shown in Figure 3.
characteristic value xt
In the following, 3 possibilities are presented to determine the
c) C o m b in a tio n o f test v a lu es + a p r io r i in form ation
required values Xm and V x:

• N o test values available T he concept o f Bayes' theorem can be used to combine a priori
knowledge (experience, judgment) with measured test values.
• Numerical test values available (e.g. several angles o f
The method presented here has been adapted from proposals by
internal friction (J>)
Tang (1 9 7 1 ). T he application o f Bayes' theorem results in a
• Combination o f the test values + a priori information combined a posterior information, which is (in a statistical sense)
superior to the a priori as well as the test results alone.

a) N o test v a lu es a v a ila b le (n = 0)

a + 4 ■b + c
(7)

Xm a + 4 •b + c
( 8)

where: a = estimated minimum value


b = most likely value Figure 2
c = estimated maximum value

T he estimation o f the values a, b and c can be based on personal

2273
c) D eter m in a tio n o f the c h a r a c te r is tic u n d ra in e d s h e a r
Numerical procedure
stren g th u sin g B a y e s ' th eo rem to c o m b in e th e in form ation
f o r o b ta in in g th e b es t es tim a te o f the m ea n (se e F ig u r e 3)
• Estim ated a priori values: xm]) V xi and S x| = x ml •Vxl
As described in 4 .2 .a)
cum3 = 6 8 kN/m2 "'I
• T est values: xm2, S » 2 and V *2 Scu3 = 7 .7 kN/m2 > cuu = 64.3 kN/m2
V c u3 = 0 .1 1 J

6 S U M M A R Y AND C O N C LU SIO N S

• Combined a posteriori information: x,„3 , S x3 and The characteristic values Xk used in today's approach (global
factor o f safety) as well as in E C 7 are for all practical purposes
equivalent and should be o f the same magnitude.
In tod ay's practice x k is selected primarily based on experience
and personal judgment. This is a well-proven practice and can, or
even should, still be employed for E C 7 Difficulties however can
arise as soon as test values are available in addition to local
experience. T he question then is: How can they be treated to
produce characteristic values or how can they be combined with
local experience? These problems mainly arise, because today the
desirable safety target is not explicitely formulated, but is more a
m atter o f intuition, experience, “feeling“ and sometimes even
economy.
E C 7 however does go one step further and presents a
B e st estimate o f the true mean: definition o f x k in qualitative and even quantitative terms. This
puts the proposed safety system on a more rational basis. X k in
E C 7 constitutes an essential part o f the safety chain governing
the entire design process The characteristic value xk therefore is
the fundamental soil properly in E C 7, from which the
= 1 : judgment/experience
corresponding design values x j are readily derived by means o f
= 2 : test values the list o f partial factors o f safety as given in E C 7, Table 2 . 1.
= 3: combined knowledge T he statistical mean value xm itself is not used in E C 7, but will
always be replaced by the corresponding characteristic value Xk.

REFEREN CES
5 EXA M PLE
Denver, H. & Ovesen, N .K . 1994. Assessment o f Characteristic
E x a m p le w ith v a lu es f r o m v an e s h e a r tests (D en ver & O vesen, Values o f Soil Parameters for Design. P roc. X IIIIC S M F E .
1 99 4 ) N ew Delhi, India.
E urocod e 7. Part 1 G e o te c h n ic a l D esign , G e n e r a l R u les. Final
T he follow ing results from 5 undrained vane shear tests (n=5) o f V ersion o f E N V 1997-1, Oct. 3. 1994, produced by CEN.
a clay till were presented: Kay, J.N . & Krizek, R .J. 1971. Estim ation o f the Mean for Soil
Properties. Int. C o n f. o n A p p lica tio n s o f S tatistics a n d
50 ; 5 0 , 6 0 ; 81 and 89 kN/m2 P r o b a b ility to S o il a n d S tru ctu ral E n g in eerin g , Hong Kong.
Lumb, P. 1974. Application o f Statistics in Soil M echanics.
All the values are in the influenced volume o f a strip footing to Chapter 3 in S o il M ech a n ics-N ew H oriz on s, ed. by I K. Lee.
be designed on the basis o f the characteristic undrained shear Orr, T 1993 U se o f Partial Factors in Eurocode 7. P a p e r
strength cuk. It is assumed that the vane strength values p r e s e n t e d a t 6th m ee tin g o f S C 7, B erlin
correspond also to the undrained shear strength (Denver & Rethati, L. 1988. Probabilistic Solutions in Geotechnics.
Ovesen, 1994). D e v e lo p m en ts in G e o te c h n ic a l E n g in e er in g 46, Elsevier,
Amsterdam.
T he estimated a priori information is given in the paper as: Schneider, H R . 1990. Die Wahl der Bodenkennwerte (in
German). A n w en du n g d e r n eu en T ragw erksn orm en d e s SIA
x„i = curai = 100 kN/m2 "I im G ru n dbau , S IA -D oku m en tation D 06 4 Zurich,
V xi = V cu , = 0 .2 5 J Scu, = 25 kN/m2 Switzerland
Tang, W .H. 1971. A Bayesian Evaluation o f Information for
Foundation Engineering Design. Int. C on f. o n A p p lica tio n s
a) D eter m in a tio n o f th e c h a r a c te r is tic u n d ra in e d s h e a r o f S ta tistics a n d P r o b a b ility to S o il a n d S tru ctu ral
stren g th f r o m th e a p r io r i in form ation on ly (se e F ig u re 3) E n g in eerin g . Hong Kong.
Terzaghi, K. 1940. Sampling, Testing and Averaging.
cun = 87.5 kN/m2 P r o c e e d in g s , P u rd u e C o n fe r e n c e on S o il M e c h a n ic s a n d its
A p p lica tio n s. Purdue University, W est Lafayette, U SA .

b) D eter m in a tio n o f th e c h a r a c te r is tic u n d ra in ed sh ear


stren g th f r o m th e v an e tests on ly ( s e e F ig u r e 3)

cUm2 = 6 6 kN/m2 'j


S Cu2 = 18.04 kN/m2 i- cuti = 57.1 kN/m2

V c u 2 = 0 .2 7 J

2274

You might also like