International Journal of Pavement Engineering

Vol. 12, No. 3, June 2011, 201214

Development of models for the interpretation of the dynamic cone penetrometer data
Melanie Bouteta1, Guy Doreb*, Jean-Pascal Bilodeaub2 and Pascal Pierreb3
a
Roche Ltee Groupe-Conseil, Quebec, Canada G1W 4Y4; bDepartment of Civil Engineering, Laval University, Quebec, Canada G1V 0A6
(Received 10 September 2008; final version received 20 April 2010)

This study deals essentially with the development of models that describe the relations between the strength properties of
soils (undrained shear strength obtained with a vane tester) on one hand, and the resilient properties (back-calculated
modulus obtained with deflection tests) on the other hand, and the penetration index values obtained with the dynamic cone
penetrometer. Laboratory tests were performed on five coarse-grained reconstituted soils. Also, field tests were performed
on 10 fine-grained soils and on each of the four subgrade soils of the Laval University Experimental Test Road Site. The
results obtained in the laboratory and in the field were compared and analysed in order to develop the relations between the
strength and the resilience for different types of soils. Using a statistical analysis with three different sophistication levels,
reliable correlations that include explanatory variables were developed.
Keywords: mathematical correlations; penetration index; modulus; strength; resilience; geotechnical properties

1. Introduction typically found in the province of Quebec (Canada) is

The dynamic cone penetrometer (DCP) is a very useful fundamental to ensure a good exploitation of the data
tool for pavements structural evaluation or for pavement obtained with the DCP. Table 1 presents a summary of the
design studies. Indeed, the DCP is often used in pavement main correlations found in the literature between the DCPI
engineering for the evaluation of the characteristics and the elastic, resilient and back-calculated modulus.
(thickness and properties) of each layer of the pavement
structure. The DCP is widely used in many countries for its
simplicity, portability, low cost and capacity to obtain data 2. Objective
allowing the rapid determination of in situ mechanical The main objective of this study is to develop models that
properties. This bearing capacity tool and test method are relate the strength properties of soils (undrained shear
also standardised (ASTM 2003). In fact, when combined strength obtained with a vane tester) on one hand, the
with correlations, the DCP allows the determination of resilient properties (back-calculated modulus obtained
basic characteristics of the pavement layers without having with deflection tests) on the other hand, and the penetration
to perform long and expensive destructive tests. index values. More specifically, the objective is to define
In the literature, there are correlations that establish the relationship between the strength and the resilience for
relationships between the dynamic cone penetration index different kinds of soils and to see if it is possible to make
(DCPI) and some parameters such as the California bearing groupings. The relationships should include parameters
ratio (CBR) and the elastic, resilient and back-calculated that take into account soil characteristics.
modulus commonly used in pavement design. Although
CBR tests are less used nowadays, the correlation that relates
directly the DCPI values to modulus is valid only for a 3. Experimental programme
limited range of conditions. This is an important fact, as the 3.1 Tests methodology and parameters considered in
American Association State Highway and Transportation models development
Officials (AASHTO 1993) suggests using the resilient The experimental programme was developed based on
modulus of soils for pavement design. Moreover, only few previous work and, among others, on a study conducted by
relationships found in the literature consider soil character- Lambert et al. (2006) on the assessment of coarse granular
istics such as Atterberg limits, water content and density. pavement foundation materials. It is divided into two main
Because there are gaps in the understanding of the parts. The first one consists in laboratory testing of five
relationship between the strength (plastic or viscoelastic coarse-grained reconstituted soils and the second one, in
domain) and the resilience (elastic domain) of soils, the field testing of 10 fine-grained soils and the four subgrade
elaboration of complete models representative of soils soils of the Laval University Experimental Test Road Site

*Corresponding author. Email: [email protected]

ISSN 1029-8436 print/ISSN 1477-268X online
q 2011 Taylor & Francis
DOI: 10.1080/10298436.2010.488727
http://www.informaworld.com
202 M. Boutet et al.

(SERUL). For each type of soil, geotechnical and

Fine and coarse grained 2 , DCPI , 60

Notes: E, elastic modulus (MPa); EFWD, retro-calculated modulus (MPa); Mr, resilient modulus (MPa); DCPI, penetration index (mm/blow); gdr, density/maximum dry density; wc, water content (%); LL, liquid limit (%); Cu, uniformity
mechanical behaviour characterisation tests were done.

Coarse grained 5.6 , DCPI , 40.0

The geotechnical characterisation tests include a particle

Fine grained 3.7 , DCPI , 66.7

size analysis and the determination of the Atterberg limits

Fine grained 10 , DCPI , 60

Tested materials and the grain density. With respect to mechanical
behaviour characterisation tests, they include the determi-
nation of the penetration index obtained with an 8 kg DCP,
the back-calculated modulus obtained with a portable
Subgrade soils
Subgrade soils
falling weight deflectometer (FWD), the undrained shear

Fine grained
strength obtained with a vane tester and the CBR. Also, the
Unknown
Unknown

water content and the density were determined in order to

know the state of the tested soils. Table 2 presents a
summary of the tests and the aimed parameters.

3.2 Laboratory tests description

The laboratory tests were performed on four coarse-

grained reconstituted soils: a poorly graded sand (SP), a

g0:674

poorly graded gravel (GP), a silty sand (SM) and a silty

dr
1:925

gravel (GM) according to the Unified Soil Classification

dr LL=wc

w20:935

System (USCS). These choices were made in order to

obtain some variability in the soil types that allows to
cr
Mr 7013.065 2040.783 ln(DCPI)

measure significant variations in the soil mechanical

20:305

properties and to obtain typical Quebecs coarse-grained

log(E) 3.65 2 1.17 log(DCPI)
log(E) 3.05 2 1.07 log(DCPI)

M r 27:86 DCPI20:114 g7:82

Correlations

and fine-grained soils. Also, these choices were influenced

EFWD 2224 DCPI 2 0.996

M r 90:68 DCPI=logcu
EFWD 338 DCPI 2 0.39

by the available soil fractions that were blended to obtain

E 17.6 (269/DCPI)0.64

such soils. As a matter of fact, natural sand was sampled at

EFWD 2191.4/DCPI

a borrow pit. The quite uniformly graded gravel fraction of

this material was also sieved by the producer. This gravel
fraction was used to blend GP and GM soils. The sand
fraction was also quite uniformly graded. Quarry fines of a
similar mineralogical composition were blended with the
sand and gravel to obtain the silty soils. For each of these
soils, four states were reconstituted: loose and saturated,
loose and drained, dense and saturated and dense and
drained. In this study, the terms saturated and drained
should to be precised. The saturated state corresponds to
phreatic conditions, the water level being adjusted near the
surface. The drained condition is created by draining
George and Uddin (2000)
Chai and Roslie (1998)a
Chai and Roslie (1998)a

the container in which the soils are compacted, lowering

Alshibli et al. (2005)

the water level at the bottom of the soils. Although a wide

Chen et al. (1999)
Authors

range of soils were tested, this drained condition creates

De Beer (1990)

Hassan (1996)b

various drained water contents because of the various

coefficient; wcr, water content/optimal water content.
Pen (1990)

water retention capacities of the tested soils. We also

considered to perform tests on clayey sand, but it was
Correlations summary.

impossible to reach the desired conditions of the

experimental programme, particularly for the compaction
Mr (psi) and DCPI (in/blow).

and the saturation of this type of soil. Therefore, the tests

DCPI (blows/300 mm).

were replaced by comparable ones in the field.

EFWD DCPI

3.2.1 Equipment description

Mr DCPI
E DCPI

The laboratory tests were performed in a large plastic

Table 1.

container with an internal diameter of 1.2 m and a height of

1.3 m. A piezometer is used to measure the water level and
b
a
 International Journal of Pavement Engineering 203

Table 2. Tests matrix and aimed parameters.

Characterisation tests
Soil type State G wL/wP rs rd max/wopt CBR DCPI EFWD Cu wc r
p pa p p p p p
4 coarse grained (laboratory) Loose saturated p p p p
Loose drained p p
Dense saturated p p p p
Dense drained p p p p p p p p
10 fine grained (field) Point A p p p p p p p p
Point B p pa p p p p p p p
4 SERUL pits (field) Point A p pa p p p p p p p
Point Ba

Notes: G, particle size analysis; wL/wP, Atterberg limits; rs, grain density; rd max/wopt, maximum dry density/optimal water content; CBR, California bearing ratio; DCPI,
penetration index; EFWD, retro-calculated modulus; Cu, undrained shear strength; wc, water content; r, density.
a
If applicable.

a drainage valve is installed at the bottom of the container. successive layers of 15 cm instead of 20 cm. The soils were
A 20-cm layer of clean gravel covered with geotextile was placed inside the container using a shovel and buckets, and
installed at the bottom of the container (Figure 1). it was levelled prior to compaction with the wood rammer
or the electric hammer.

3.2.2 Sample preparation

When the fine particle content of the tested soil was under 3.2.3 Protocol (saturated state)
12%, the container was filled in successive layers of 20 cm To get to the saturated state, the container was filled with
slightly densified with a wood rammer until the soil water in order to get full immersion of the soil sample. A
thickness was 1 m (^ 10 cm). When the fine particle minimum rest period of 24 h was required to ensure a good
content of the tested soil was over 12%, the container was saturation of the compacted soil. It should be noted that the
filled in successive layers of 20 cm densified with a water level was controlled with the external piezometer that
circular plate of 10 cm diameter fixed to an electric is installed on the side of the container. Prior to testing, the
hammer. This was made following a detailed compaction water level was lowered 75 mm under the surface in order to
procedure including loading pattern and application time. obtain a thin layer of stable soil to facilitate testing. The
To obtain a denser state, the container was filled in FWD test was performed using a 300 mm diameter plate
with the geophone in contact with the soil surface. The
device was positioned to ensure good surface contact and
the test was done at the centre of the container. For each
applied stress levels (30, 50, 70 and 100 kPa), four drops
were applied, the first one being performed in order to
ensure good plate contact. The average of the results
obtained with the three subsequent drops was calculated.
The DCP test was performed at three distinct places in the
container (Figure 2). The DCP tests were performed
approximately at the middle of the radius (Figure 2) and the
FWD tests were performed at the container centre to
prevent any influence of the deflection tests on the
penetration values. Preliminary tests concluded that there
was no significant changes in the results if the FWD tests
were performed prior to DCP tests and vice versa.

3.2.4 Protocol (drained state)

To get to the drained state, the valve of the container was
opened. The soil was considered drained when the water
level did not increase in the piezometer when the valve is
closed. To minimise evaporation, a polythene cover was
Figure 1. Container for the laboratory tests. installed over the container during drainage. It should be
204 M. Boutet et al.

3.3.1 Protocol
Firstly, the FWD test was performed using a 300-mm
diameter plate with the geophone in contact with the soil
surface. The device was positioned in order to ensure good
surface contact and the test was done at the centre of the first
trench referred to as the lower point. For each applied
stress levels (30, 50, 70 and 100 kPa), four drops were made,
the first one being performed in order to ensure good plate
contact. The average of the results obtained with the three
subsequent drops was calculated. The DCP test was
performed at three distinct places in the trench (Figure 3).
For the fine-grained soil sites, three vane tests were done
according to NQ 2501-200 (NQ 1988). For each of them,
the undrained shear strength was determined at three
different depths: 200, 400 and 600 mm. It should be noted
that this test was not performed at the SERUL because the
Figure 2. Location of the laboratory tests. subgrade soils are mostly coarse grained. In order to get
accurate characterisation information, the density and the
noted that the tests performed on the drained soil were the water content had to be determined after the FWD and DCP
same as those performed on the saturated soil except for tests. These were determined with the sand cone test
the position of the DCP tests (Figure 2). As a matter of according to ASTM D 1556-90 (ASTM 1996) at the
fact, considering the saturated state and the drained state, position of the FWD test and at two other distinct places
each DCP test was performed at approximately 50 cm (Figure 3). Finally, samples were collected in order to
apart from another DCP test to ensure that there is no perform geotechnical characterisation tests. This protocol
influence between each test. In order to determine the state was repeated for the second trench referred to as the upper
of the soil, the density and the water content had to be point that is slightly shallower than the lower point.
determined at different levels in the container just after the
FWD and DCP tests. The soil density was determined with
the sand cone test according to ASTM D 1556-90 (ASTM 3.4 Geotechnical characterisation tests description
1996) at the surface, at the position of the FWD test and at The objective of the geotechnical characterisation is to
mid ray, then at 30 and 60 cm under the surface, always assess the physical properties of the soils tested in the
between the positions of the DCP tests to make sure that laboratory and in the field. Depending on the type of soil
the soil was undisturbed. As for the water content, it was considered, various tests had to be carried out. For all soils,
determined according to ASTM D 2216-98 (ASTM 1998) the following tests were done: the particle size analysis
at the surface and at each 15-cm depth. according to BNQ 2560-040 (BNQ 1982) and the
determination of the density and the absorptivity of the
coarse aggregate according to BNQ 2560-067 (BNQ
3.3 Field tests description
The field tests were done on 10 sites covering a wide range
of fine-grained soils representative of the geological
context of the province of Quebec and on each of the
SERUL subgrade soils. For each site, the tests were
performed at two distinct places named upper point
(lower theoretical water content) and lower point (higher
theoretical water content) in a trench dug to the subgrade
soil level using a backhoe. In the case of the SERUL
subgrade soils, the tests were performed only at one depth
for each of the four different subgrade soils. However,
because the laboratory tests planned on clayey sand could
not be performed, the tests on the first two SERUL
subgrade soils had to be performed a second time at a
greater depth. Effectively, these soils were interesting
replacement test sites because they are well-documented
SC-type soils (clayey sands). Figure 3. Location of the field tests.
 International Journal of Pavement Engineering 205

Table 3. Ranges of dependent and probable explanatory variables for fine-grained soils.

Soil type Variable Attribute, symbol Range

a
Fine grained Dependent Back-calculated modulus (MPa), EFWD 14.9 231.4
Undrained shear strength (kPa), Cu 22.6 147.7
Independent Penetration index (mm/blow), DCPI 9.8 200.0
Water content (%), wc 18.0 71.9
Density (mg/m3), r 0.85 1.95
Dry density (mg/m3), rd 0.61 1.64
Per cent passing # 200 sieve (%), # 200 61.5 99.2
Liquid limit (%), wL 19.7 59.9
Plasticity limit (%), wp 12.8 25.4
Plasticity index (%), IP 6.7 36.1
Liquid index, IL 0.2 1.6
a
EFWD values obtained at 50 kPa stress.

1983). If the soil tested was characterised by less than 12% shallower limit was selected considering the adjustment
of fines, the determination of the density and the depth of the DCP, which is about 200 mm. The deeper limit
absorptivity of the fine aggregate according to BNQ was determined using the Boussinesq model, considering
2560-065 (BNQ 1989) was done. On the opposite, if the the approximate depth of influence of a load uniformly
soil tested was characterised by more than 12% of fines, distributed on a flexible plate. This model allows the
the particle size analysis by sedimentation according to determination of vertical stress as a function of depth. For
NQ 2501-025 (NQ 1987), the determination of the relative this project, the depth of influence of the FWD load was
density according to CAN/BNQ 2501-070 (CAN/BNQ determined based on a 10% residual vertical stress level
1986a) and the determination of the liquid and plastic (sz/s0 100). Thus, considering a 0.15 m radius plate
limits according to CAN/BNQ 2501-092 (CAN/BNQ (FWD test), a residual stress level of 10% is obtained for a
1986b) were done. Finally, the following tests were done depth of about 600 mm. The DCP analysis was made up to
for the coarse-grained soils: the modified Proctor test that specific depth for this reason. It should be noted that the
according to BNQ 2501-255 (BNQ 1986) and the CBR test adjustment depth of the DCP was higher than 200 mm in
according to ASTM D 1883-92 (ASTM 1992). some cases. This was observed particularly when the initial
penetration depth was over 100 mm. The DCPI was then
determined from a lower limit higher than 200 mm.
4. Results The stress level of 50 kPa was selected for the
Tables 3 and 4 list the range of dependent and probable determination of EFWD. The stress level was selected based
explanatory variables for the tested fine and coarse-grained on a theoretical assessment of vertical stress transmitted to
soils. In addition, Table 5 presents a summary of other the subgrade soil under a typical low-volume pavement.
important geotechnical properties of the tested soils as well The calculation was made using WINJULEA software for
as the compaction state obtained, expressed as rd/rd max two different types of subgrade soils: silty and clayey
values. With respect to DCPI values, it is worth noting that gravels (GM-GC) and silty and clayey sands (SM-SC) and
they were evaluated between 200 and 600 mm depth. The two loading conditions: 700 kPa under single tyre and

Table 4. Ranges of dependent and probable explanatory variables for coarse-grained soils.

Soil type Variable Attribute, symbol Range

Coarse grained Dependent Back-calculated modulusa (MPa), EFWD 11.2 187.8
Independent Penetration index (mm/blow), DCPI 13.4 250.0
Water content (%), wc 3.8 28.9
Density (mg/m3), r 1.55 2.42
Dry density (mg/m3), rd 1.34 2.26
Per cent passing # 200 sieve (%), # 200 2.1 44.4
Liquid limitb (%), wL 20.6 28.3
Plasticity limitb (%), wp 15.2 21.9
Plasticity indexb (%), IP 2.1 8.4
a
EFWD values obtained at 50 kPa stress.
b
For the coarse-grained soils characterised by 12% or more fines.
206 M. Boutet et al.

Table 5. Geotechnical properties of coarse-grained soils.

Modified Proctor test,

max. dry density (mg/m3)/
Soil sample optimum water content (%) rd/rd max (%) CBRa (%)
Reconstituted soils (laboratory)
SP 1.90/10.5 79/91 28.8/34.4
GP 2.17/5.2 88/93 203.5/256.1
SM 2.14/6.7 83/93 45.5/57.3
GM 2.28/4.3 78/99 92.1/105.7
SERUL pits (field)
Pit 1 (SC)
Point A 2.11/8.2 74 26.9/32.4
Point B 2.09/7.7 80 74.9/75.5
Pit 2 (SC)
Point A 1.99/10.2 71 40.8/40.7
Point B 1.98/9.8 71 61.4/57.5
Pit 3 (SM)
Point A 2.20/6.4 89 48.4/67.9
Pit 4 (SP)
Point A 2.14/8.9 93 130.4/160.4
a
CBR, California bearing ratio (at 2.54 mm/5.08 mm penetration).

under dual tyres. The results of the analysis are given in where e is the void index, rs is the grain density, rw is the
Table 6. WINJULEA was also used to determine the effect density of water and Sr is the saturation level. In addition,
of using a FWD plate diameter of 300 mm on a sample for simplification purposes, the following assumptions
having a diameter of 1200 mm and a depth of 1300 mm. were made:
This was done to investigate the possible boundary . The void ratio is considered constant with the
condition effects. The analysis suggested that all along the
variation of water content.
container wall and at its bottom, the calculated . The soil saturation is 100% for the saturated state.
deformations were always 10% or lower than the
calculated deformation under the plate. In addition, it For the saturated state, the 100% saturation assump-
should be noted that the container walls are smooth and tion is made because it was not possible to obtain soil
can accommodate, at least partially, vertical movements. samples for water content measurements. As a matter of
On the basis of these observations, it is suggested that fact, it was suitable to keep soils integrity for the drained-
boundary effects may be negligible in this study. condition tests. For the drained-test conditions, water
With respect to the values of water content (w), density content measurements were taken when the soil was
(r) and dry density (rd) for the saturated states of the four removed from the container to compute the actual
coarse-grained laboratory reconstituted soils, these were saturation degree.
determined with the following equations:
rs
e 2 1; 1 5. Development of mathematical models
rd
In order to develop models that relate the strength
rw Sr e properties of soils (undrained shear strength) on one hand,
w ; 2 the resilient properties (back-calculated modulus) on the
rs
other hand, and the penetration index values obtained with a
r rd 1 w; 3 DCP, a statistical analysis with three different sophisti-
cation levels was done. The first level aims at obtaining a
Table 6. Winjulea analysis. direct relationship between the two variables, i.e. without
taking into account the classification of the soils and the
Vertical stress (kPa) on subgrade soil
geotechnical characterisation test results. The first level
Soil type Single tyre Dual tyre models are meant to be applied when minimal information
GM, GC 50.7 44.7 is available on the soils tested (i.e. fine grained or coarse
SM, SC 38.6 34.9 grained). The purpose of the second level is to establish the
same type of relationship, but specific to soil classes when
 International Journal of Pavement Engineering 207

the information is available. As for the third level, the be obtained during the realisation of the field tests because of
objective is to improve the level of correlation by including the physical limitations of the vane test equipment.
soil properties easily obtained from common geotechnical
characterisation tests. For each level, the model quality is
evaluated with the root mean square error (RMSE) and the 5.1.1 Level 1
determination coefficient (R 2). In the development of the For the development of the correlation, five models were
models, five basic principles were followed: investigated: linear, power, exponential, logarithmic and
. To obtain a minimal RMSE, i.e. a minimal difference log log. For each one of them, the RMSE and R 2 values
between the predicted and the observed values, it are evaluated by a regression analysis. Moreover, the data
should be noted that a model that provides a slightly normality is verified with the Kolmogorov-Smirnov test.
higher RMSE but that takes into account more The comparison of the results has shown that the best-fit
relevant parameters was sometimes privileged. model is the exponential model. Thus, the suggested
. The larger the coefficient of determination, R 2, the relation for level 1 is
better the model is. logC u 20:0042 DCPI 2:10;
. The model should allow the normality of the data so 4
that parametrical tests can be performed. Here, the RMSE 26; R 2 0:78; N 13:
normality was confirmed with the Kolmogorov-
Smirnov test.
. The significance of the relation between the dependent 5.1.2 Level 2
variable and the explanatory variables included in the As it can be observed in Figure 4, the soils tested can be
model was confirmed with a Fisher F-test in order to subdivided in two categories. The first includes CH-type
compare the model quality with and without soils (inorganic clays with high plasticity) and the second,
explanatory variables. CL-type soils (inorganic clays with low to medium
. The significations of the individual coefficients of the plasticity) and ML-CL-type soils (clayed silts and silty
model was confirmed with a Student t-test in order to clays). In order to confirm the relevance of this subdivision
verify if the confidence interval of each coefficient can in the development of the models, two tests must be done.
or cannot contain the zero value. These tests compare the slope and the intercept point
values for the two types of soils with a covariance analysis
using the XLStat software. More specifically, a subdivi-
5.1 Relationship between undrained shear strength and sion is justified if
penetration index
Px q . 0:05 and=or Pq . 0:05; 5
Data obtained for the fine-grained soil sites are presented in
Figure 4. It should be noted that some values had to be where x is the quantitative explanatory variable and q is the
rejected from the analysis. In fact, some observations were qualitative explanatory variable.
related to organic soils with low to medium plasticity (OL) as Because the P(log(DCPI)*SoilType) 0.1% and that
well as SM-SC, which are not considered for this part of the P(SoilType) 4.1%, the soils tested are subdivided into
study. Also, some undrained shear strength values could not two categories (distinct slopes and intercept points). Thus,
the suggested relation for level 2 is

logCu a0 logDCPI 2:50 g;

CH soils : a0 21:19 and g 1:54;
CL and ML-CL soils : a0 20:43 and g 0;
RMSE 11; R 2 0:92; N 13:
6

It should be noted that the presented level 2 model is

log log instead of exponential because it was the best-fit
model for both types of soils.

5.1.3 Level 3
Figure 4. Undrained shear strength (Cu) vs. penetration index In order to select the physical parameters that are the most
(DCPI). likely to decrease the RMSE and to increase the R 2,
208 M. Boutet et al.

Table 7. Correlation matrix of explanatory variables for CH-type soils.

w r rd wL wP %F IP IL log(Cu) log(DCPI)
w 1 2 0.98 20.99 0.96 2 0.79 20.24 0.96 0.99 2 0.87 0.84
r 1 1.00 2 0.96 0.83 0.29 2 0.97 2 0.98 0.89 2 0.85
rd 1 2 0.95 0.84 0.32 2 0.97 2 0.99 0.91 2 0.87
wL 1 2 0.77 20.16 0.99 0.94 2 0.84 0.78
wP 1 0.76 2 0.85 2 0.83 0.99 2 0.98
%F 1 2 0.29 2 0.32 0.66 2 0.72
IP 1 0.95 2 0.91 0.85
IL 1 2 0.89 0.88
log(Cu) 1 2 0.98
log(DCPI) 1

correlation matrix values are calculated with the relevant individual coefficients, it was tested with a Student t-test.
explanatory variables for each type of soil (Tables 7 and 8). At a confidence level of 95%, all the coefficients
By carrying out optimisation tests while including are significant with jt*j values higher than the critical
combinations of the most relevant explanatory variables values t(n2).
according to the correlation matrix, it is possible to Figure 5 presents the predicted and measured Cu values
improve the quality of the model. In fact, optimised results for the level 3 model (Equation (7)). With points clustering
are obtained by including in the relationships a new along the line of equality, the evidence is sufficient to
explanatory variable, which combines the plasticity index state that the addition of parameters, which take into
(IP) and the dry density (rd). Thus, the suggested relation account soil properties, has allowed the development of a
for level 3 is high-predictability model.

1=2
logCu a0 I P rd 1=2 logDCPI 2:39 g;
5.2 Relationship between back-calculated modulus and
CH soils : a0 20:20 and g 1:51;
penetration index
CL and ML-CL soils : a0 20:10 and g 0; Figure 6 presents the data obtained for the fine-grained soil
RMSE 7:8; R 2 0:95; N 13: sites, whereas Figure 7 presents data obtained for the SERUL
7 subgrade soils and the coarse-grained laboratory reconsti-
tuted soils. Because the behaviour of the fine-grained soils
Table 9 presents the results of the parametric tests differs significantly from that of the coarse-grained soils,
performed during the development of the models. To relationships are developed for each of these groups. It
validate the significance of the relationship between Cu should be noted that some values had to be rejected from the
and all of the explanatory variables included in the models, analysis. In fact, for the fine-grained soils, some observations
the F-test was conducted. With F* values much higher were related to organic soils with low to medium plasticity
than the critical values F(0.95, n1, n2), it can be concluded (OL), which are not part of the study. Also, the results of the
that a strong relationship between the undrained shear inorganic silt with low to medium plasticity (ML) did not
strength and the selected explanatory variables exists. Note seem representative of what is usually found in the field.
that n1 is the model degrees of freedom and n2 is the error With respect to the coarse-grained soils, some penetration
degrees of freedom. With respect to the significance of the index values were out of range according to the main range of

Table 8. Correlation matrix of explanatory variables for CL and ML-CL-type soils.

w r rd wL wP %F IP IL log(Cu) log(DCPI)
w 1 2 0.93 20.96 0.94 0.88 0.63 0.96 0.23 2 0.66 0.47
r 1 1.00 2 0.87 2 0.79 20.46 2 0.90 2 0.23 0.59 2 0.46
rd 1 2 0.88 2 0.81 20.52 2 0.91 2 0.26 0.59 2 0.43
wL 1 0.98 0.68 0.99 2 0.10 2 0.66 0.55
wP 1 0.74 0.94 2 0.21 2 0.56 0.46
%F 1 0.62 2 0.05 2 0.07 2 0.19
IP 1 2 0.03 2 0.73 0.59
IL 1 0.06 2 0.25
log(Cu) 1 2 0.88
log(DCPI) 1
 International Journal of Pavement Engineering 209

Table 9. Results of the parametric tests for the undrained shear strength (Cu).

Level Soil type Coefficient Value jt*j F*

a c
1 All types a0 2 0.0042 6.17 38.11e
a1 2.10 33.38c
2b All types a1 2.50 18.91d 35.08f
b 1
CH a0; g 21.19; 1.54 8.38d; 4.92d
CL and ML-CL 20.43; 0 4.81d
3b All types a1 2.39 28.10d 56.31f
1=2
b I P rd 1=2
CH a0; g 20.20; 1.51 10.56d; 6.57d
CL and ML-CL 20.10; 0 6.25d
a
Model: log(Cu) a0 DCPI a1.
b
Model: log(Cu) a0 b log(DCPI) a1 g.
c
t0.05 (11) 1.80.
d
t0.05 (9) 1.83.
e
F0.05 (1,11) 4.84.
f
F0.05 (3,9) 3.86.

values obtained and others appeared to be influenced by the Coarse-grained soils:

presence of recycled asphalt concrete in the soil of the
logEFWD 20:62 logDCPI 2:56;
test site.
9
RMSE 45; R 2 0:42; N 19:
5.2.1 Level 1
As for the shear strength relationships, five models
were investigated for the development of the correlations: 5.2.2 Level 2
linear, power, exponential, logarithmic and log log. For As it can be observed in Figures 6 and 7, the soils tested
each one of them, the values of RMSE and R 2 are can be subdivided into two categories. For the fine-grained
evaluated by a regression analysis. Moreover, the normal- soils, the first group includes CH-type soils (inorganic
ity of the data is verified with the Kolmogorov-Smirnov clays with high plasticity) and the second, CL-type soils
test. The comparison of the results has shown that the best- (inorganic clays with low to medium plasticity). For the
fit model of the fine and coarse-grained soils is the log log coarse-grained soils, the first group takes into account SP-
model. Thus, the suggested relationships for level 1 are and GP-type soils (poor graded sands and gravels) and the
Fine-grained soils: second, SC-, SM-, SM-SC- and GM-type soils (clayey
sands, SMs and GMs). In order to confirm the relevance of
logEFWD 20:45 logDCPI 2:52; these subdivisions in the development of the models, the
8 same tests as those described in Section 5.1.2 were done.
RMSE 54; R 2 0:33; N 13: For the fine-grained soils, P(log(DCPI) * Soil

Figure 6. Modulus (EFWD) vs. penetration index (DCPI) for

Figure 5. Measured vs. predicted undrained shear strength (Cu). fine-grained soils.
210 M. Boutet et al.

optimised results are obtained by including in the

relationships the density (r) and the plasticity index (IP).
Thus, the suggested relationships for level 3 are
Fine-grained soils:

logEFWD a0 b logDCPI 2:82 g;

1=4
CH soils : a0 21:17; b I L rd and g 1:08;
CL soils : a0 22:78; b I P rd 21=2 and g 0;
RMSE 12; R 2 0:95; N 13:
12
Coarse-grained soils:

Figure 7. Modulus (EFWD) vs. penetration index (DCPI) for logEFWD a0 b logDCPI 6:04 g;
coarse-grained soils.
SP and GP soils : a0 20:26; b r and g 23:58;
SM; SC; SM-SC and GM soils : a0 23:47;
Type) 56.6% and P(SoilType) 0.1%, so the soils
b I 20:018
P and g 0;
tested are subdivided into two categories (same slopes
but distinct intercept points). For the coarse-grained RMSE 9:8; R 2 0:88; N 19:
soils, P(log(DCPI) *SoilType) 0.01% and P(Soil 13
Type) 0.2%, so the soils tested are also subdivided in
two categories (distinct slopes and intercept points). Thus, The use of the density for the coarse-grained soils may
the suggested relations for level 2 are be unusual. However, it is justified by the correlation
Fine-grained soils: matrix. As the density is associated with dry density and
water content, it seems that the inclusion of density in the
log EFWD 20:84 logDCPI 2:94 g; model may statistically explain the effect of dry density
and water content in a combined manner. However,
CH soils : g 0:59;
10 because it is more convenient for the coarse-grained soils
CL soils : g 0; to use geotechnical characteristics more easily obtained in
RMSE 27; R 2 0:82; N 13: the field, a relationship that has a slightly higher RMSE but
that takes into account water content was developed:
Coarse-grained soils: Coarse-grained soils:

logEFWD a0 b logDCPI 6:65 g;

logEFWD a0 logDCPI 6:01 g;
SP and GP soils : a0 20:013; b w and g 24:89;
SP and GP soils : a0 20:65 and g 23:30;
SM; SC; SM-SC and GM soils : a0 24:04;
SM; SC; SM-SC and GM soils : a0 23:37 and
b w 20:016 and g 0;
g 0; RMSE 13; R 2 0:88; N 19: (11
RMSE 10:4; R 2 0:85; N 19:
14
5.2.3 Level 3
In order to select the physical parameters that are the most In order to validate the significance of the relationship
likely to decrease the RMSE and to increase the R 2, between EFWD and all of the explanatory variables included
correlation matrix values similar to those presented in in the models, the F-test was conducted. With F* values
Section 5.1.3 are calculated with the relevant explanatory much higher than the critical values F(0.95, n1, n2), it
variables for each type of soil. can be concluded that a strong relationship between the
By carrying out optimisation tests while including back-calculated modulus and the selected explanatory
combinations of the most relevant explanatory variables variables exists. With respect to the significance of the
according to the correlation matrix, it is possible to individual coefficients, it was tested with a Student t-test. At
improve the quality of the model. For the fine-grained a confidence level of 95%, all the coefficients are significant
soils, optimised results are obtained by including in the with jt*j values higher than the critical values t(n2).
relationships new explanatory variables which combine Figure 8 presents the predicted and measured EFWD
the liquid index (IL), the plasticity index (IP) and the dry values for the level 3 (Equations (12) (14)). With points
density (rd). With respect to the coarse-grained soils, clustering along the line of equality, the evidence is
 International Journal of Pavement Engineering 211

measurements instead of the software Mr default value,

which is too low, creates a total benefit of 17,340 $ per
kilometre. On the contrary, for the SP-type soils, the
software Mr default value overestimates the modulus, and
a supplementary cost of 13,980 $ per km of pavement
structure is needed to ensure a safe and adequate design.
With respect to the models precision in terms of pavement
cost variation per kilometre (Table 11), it increases with
the model level and thus it demonstrates the relevance of
using the higher level when possible.

7. Discussion
Figure 8. Measured vs. predicted modulus (EFWD). The DCP is a very useful tool which allows determining
many pavement layer characteristics when combined with
proper interpretation models. In this study, models are
sufficient to state that the addition of parameters, which developed in order to ensure a good exploitation of the
take into account soil properties, has allowed the data obtained with the DCP in the geological context of
development of a high-predictability model. the Quebec province in Canada. The models describe the
relationship between the strength properties of soils
(undrained shear strength obtained with a vane tester) on
6. Cost-benefit analysis one hand, the resilient properties (back-calculated
In order to quantify the benefits of the developed models modulus obtained with deflection tests) on the other
between the back-calculated modulus (obtained with hand, and the penetration index values for various soils.
deflection tests) and the penetration index values, a cost- The models that describe the relationship between the
benefit analysis was carried out using the Quebec Ministry undrained shear strength and the penetration index values
of Transportation (MTQ) flexible pavement design soft- are derived for the following data range: 10 , DCPI ,
ware CHAUSSEE2. By carrying out structural optimis- 200. Depending on the desired precision and on the
ation tests, the total material cost of a typical local information available on the soil properties, three levels of
pavement is evaluated in $/m2 for two different subgrade relationships are proposed. With respect to the models that
soil types: CH (inorganic clays with high plasticity) and SP describe the relationship between the back-calculated
(poor graded sands). For each one of them, the cost modulus and the penetration index values, they are derived
variation per kilometre of the pavement structure for the following data ranges: 10 , DCPI , 200 (fine-
associated with the variation of the subgrade resilient grained soils) and 10 , DCPI , 250 (coarse-grained
modulus value (Mr) is evaluated. This was done by soils). For these models, only the relationships at levels
comparing the results obtained with the software Mr 2 and 3 are recommended. The level 1 relationship is
default value and the tested soils mean Mr value (Table 10). considered inaccurate and it may lead to significant
Also, the precision of the models developed in Section 5.2 estimation errors.
is interpreted in terms of pavement cost variation per With respect to the relationship between the strength
kilometre in Table 11. properties of soils and the penetration index values, high
As it can be observed in Table 10, the variation of the predictability models are obtained for the three interpret-
subgrade resilient modulus from the default value to the ation levels. In fact, the addition of parameters which take
tested soils mean value causes a non-negligible pavement into account soil properties is not essential to obtain a low
cost variation per kilometre. In fact, for the CH type soils, error. Concerning the relationship between the resilient
the use of subgrade Mr value estimated from DCP properties of soils and the penetration index values, it is

Table 10. Cost-benefit analysis.

Subgrade type Subgrade resilient modulus (MPa) Pavement cost variation per kma ($)
CH (IL , 0.9) Default value 31 17,340
Tested soils mean value 99
SP Default value 82 13,980
Tested soils mean value 30
a
Pavement width 6 m.
212 M. Boutet et al.

Table 11. Models precision in terms of pavement cost analysis. With sufficient information, this kind of analysis
variation. can be performed prior to the study or after the study to
verify the validity of the obtained results.
Pavement cost
Subgrade type Model variation per kma ($) In statistical terms, Cohen (1988) defines the power as
the probability of rejecting a null hypothesis when the
CH (IL , 0.9) Level 1 16,140
Level 2 9060 alternate is true, or as the probability of rejection of an
Level 3 1740 actually false null hypothesis. In other words, the power is
SP Level 1 96,300 the ability to detect an effect, given that this particular
Level 2 8700 effect exists in reality. This concept is associated with
Level 3 4860 the statistical error type. Two main error types may be
a
Pavement width 6 m.
experienced in statistical analysis, which are types 1 and 2
errors. The type 1 error is the significance level a, which is
the odd of finding a relationship although there is none in
essential to add soil characteristics to obtain a low error. In reality. This is used to compute the confidence level 1 2 a,
which was set to the traditional value of 0.95 in this study
order to identify the key parameters likely to improve the
as previously mentioned. Type 2 error is the odd, b, of not
models, a regression analysis was performed using the
finding a relationship although there is actually one. The
variance and the covariance. For the fine-grained soils,
power is expressed as 1 2 b, and refers to the odd of
these parameters are the dry density (rd), the plasticity
saying there is a relationship and that it actually exists.
index (IP) and the liquidity index (IL), whereas for the
A statistically powerful model is usually set at a level of
coarse-grained soils, these are the density (r) and the
0.8 (b 0.2).
plasticity index (IP) or the water content (w).
In a power analysis, the adequate sample size
This research effort was important, as several
calculation for regression depends on four basic
limitations were identified for the available models
parameters, which are the significance level a, the number
found in the literature and the performed study allowed
of predictors included in the model, the desired statistical
extending these limitations. The research allowed devel-
power and the effect size. The sample size is the parameter
oping statistical models which enable calculating the that has the more pronounced effect on the statistical
subgrade soils modulus with an accessible, inexpensive, power. The effect size is a measure of the strength of the
rapid and simple test like dynamic cone penetration. In relationship between two variables and is associated with
addition, this tool allows obtaining soils bearing capacity, the degree to which a relationship actually exists. Several
as well as a characterisation of layer thicknesses, if a ways of computing the effect size are suggested in the
change of layer occurs in the analysed subgrade soil. This literature (Cohen 1988). For our study, the most
study is original as the estimation of the modulus for fine- appropriate effect size-related measurement to be used in
grained and coarse-grained soils can be based on soils state the regression context f 2 is proposed by Cohen (1988) and
and physical properties. This was important to consider, is expressed as
because such an approach for the model development
allows differentiating soils in a more precised manner and R2
to consider the effect of important parameters such as f2 ; 15
1 2 R2
water content and density.
Because of the significant work load associated with in which R 2 is the determination coefficient. Fundamen-
the data collection on the field and in the laboratory, it was tally, f 2 will increase with increasing R 2. The sample size
not possible to develop the models with large number of calculation is based on the determination of the Fisher F-
measurements. As a matter of fact, soils blending and distribution probability density function, the non-central-
compaction, 1.5 m3 container filling, as well as the ity parameter of the F-distribution, the non-central F-
excavation work in the field, were significant tasks to distribution probability density function and the normal
perform. Therefore, the study was designed in order to curve cumulative distribution function. More details on the
collect information on various soil types encountered in procedure for sample size determination using the
Quebec. This goal was actually achieved because eight statistical power can be found in Cohen (1988). Never-
coarse-grained soils and ten fine-grained soils were tested theless, knowing the significance level (a 0.05), the
under various and controlled test conditions. However, power level (b 0.8), the number of predictors in the
it is more than relevant to verify if the number of considered equation and the effect size calculated from the
measurements included in all the proposed models is obtained R 2, the sample size that would be appropriate for
statistically significant to positively conclude on the the development of the proposed equations can be
models reliability. In regression statistical sciences, obtained. Table 12 summarises the details of this analysis
sample size analysis can be performed using a power that allows evaluating the appropriate sample size given
 International Journal of Pavement Engineering 213

Table 12. Sample size evaluation for the proposed models.

Equation f2 N used in the model Appropriate N

(4) 3.5454 13 6 OK
(6) CH 11.5 5 4 OK
(6) CL and ML-CL 11.5 8 4 OK
(7) CH 19 5 6
(7) CL and ML-CL 19 8 6 OK
(8) 0.4925 13 18
(9) 0.7241 19 13 OK
(10) CH 4.5556 5 5 OK
(10) CL 4.5556 8 5 OK
(11) SP and GP 7.3333 8 5 OK
(11) SM, SC, SM-SC and GM 7.3333 11 5 OK
(12) CH 19 5 6
(12) CL 19 8 6 OK
(13) SP and GP 7.3333 8 7 OK
(13) SM, SC, SM-SC and GM 7.3333 11 7 OK
(14) SP and GP 5.6667 8 8 OK
(14) SM, SC, SM-SC and GM 5.6667 11 8 OK

the obtained results and the proposed models. As it can be or DCPI Cu specific to soil classification and relationships
observed, in a general manner, the experiment was including soil properties as explanatory variables. Minimal
adequately designed and the proposed models for each information is needed on the tested soil for the first level.
sophistication level met appropriate standards in terms of For the second level, the soil classification is required,
number of measurements for a statistical power of 80%. whereas parameters which take into account soil proper-
However, three models do not meet the appropriate ties such as water content, soil consistency or density are
number of data that should be included in the regression needed for the third level.
analysis. Two of them are models proposed for CH soils The prediction error of the DCPI Cu relationships
(Equations (7) and (12)), for which only five reliable decreases from 33 to 10% with the addition of
measurements were obtained. The third one (Equation (8)) explanatory variables. With respect to the DCPI EFWD
would necessitate a higher number of data because of the relationships, the prediction error decreases from 70 to
low determination coefficient obtained for this level 1 16% for the fine-grained soils and from 75 to 16% for
model. However, as part of the recommendations the coarse-grained soils with the addition of explanatory
suggested in this paper, level 1 relationships are not variables.
recommended for the estimation of the EFWD from The models developed are applicable for almost all
penetration index data because of the higher estimation types of soils found in the geological context of the
error. This is particularly true for fine-grained soils, where Quebec province in Canada. It is recommended to use the
the sample size was found to be lower than the appropriate higher level models when possible. In fact, the first level
calculated sample size. Nevertheless, these three equations has to be used only if the explanatory variables are not
still may be used with caution and good engineering available or as a first approximation. The models are not
judgement. Even if most of the proposed models are recommended for coarse-grained soils with a penetration
statistically valid, further researches should be focused on index below 10 mm/blow because these values are
improving the models by the collection of supplementary beyond the confidence interval of the proposed relations.
data. Among other things, this approach for the proposed
models should allow increasing the predictability, as well
as increasing the applicability in terms of soil types, soil Acknowledgements
conditions and DCPI range of values. The authors gratefully acknowledge the Quebec Ministry of
Transportation for their financial and technical support for this
project. In addition, the authors wish to acknowledge the research
assistant who contributed to this research effort.
8. Conclusion
This study has led to the development of reliable models
between the penetration index and the undrained shear Notes
strength on one hand and the back-calculated modulus on 1. Email: [email protected].
the other hand. The relationships are developed at three 2. Email: [email protected].
different levels including direct relationships DCPI EFWD 3. Email: [email protected].
214 M. Boutet et al.

References Chai, G. and Roslie, N., 1998. The structural response and
AASHTO, 1993. Guide for design of pavement structures. behavior prediction of subgrade soils using falling weight
Washington, DC. deflectometer in pavement construction. Proceedings of the
Alshibli, K.A., Abu-Farsakh, M. and Seyman, E., 2005. 3rd international conference on road and airfield pavement
Laboratory evaluation of the geogauge and light falling technology, Beijing, China, 1481 1489.
weight deflectometer as construction control tools. Journal of Chen, J., Hossain, M. and LaTorella, T.M., 1999. Use of falling
Material in Civil Engineering, 17 (5), 560 569. weight deflectometer and dynamic cone penetrometer in
ASTM, 1992. Standard test method for CBR (California Bearing pavement evaluation. Transportation Research Record:
Ratio) of Laboratory-Compacted Soils. ASTM D 1883-92. Journal of the Transportation Research Board, 1655,
ASTM, 1996. Standard test method for density and unit weight of 145 151.
soil in place by the sand-cone method. ASTM D 1556-90. Cohen, J., 1988. Statistical power analysis for the behavioral
ASTM, 1998. Standard test method for laboratory determination sciences. Mahwah, NJ: Lawrence Erlbaum Publishers.
of water (moisture) content of soil and rock by mass. ASTM De Beer, M., 1990. Use of dynamic cone penetrometer (DCP) in
D 2216-98. the design of road structures. In: G.E. Blight, et al. eds.
ASTM, 2003. Standard test method for use of dynamic cone Geotechnics in African environment. Rotterdam: Balkema,
penetrometer in shallow pavement applications. ASTM D 167 176.
6951-03. George, K.P. and Uddin, W., 2000. Subgrade characterization for
BNQ, 1982. Granulats analyse granulometrique par tamisage. highway pavement design. Final report, University of
Bureau de normalisation du Quebec (Quebec, Que.), BNQ Mississippi.
2560-040. Hassan, A., 1996. The effect of material parameters on dynamic
BNQ, 1983. Granulats determination de la densite et de cone penetrometer results for fine-grained soils and granular
labsorptivite du gros granulat. Bureau de normalisation du materials. Thesis (PhD). University of Oklahoma State.
Quebec (Quebec, Que.), BNQ 2560-067. Lambert, J.P., Fleming, P.R. and Frost, M.W., 2006. Laboratory
BNQ, 1986. Sols determination de la relation teneur en eau
assessment of coarse granular foundation materials. Trans-
masse volumique essai proctor modifie. Bureau de
portation Research Record: Journal of the Transportation
normalisation du Quebec (Quebec, Que.), BNQ 2501-255.
BNQ, 1989. Granulats determination de la densite et de Research Board, 1952, 58 67.
labsorptivite du granulat fin. Bureau de normalisation du NQ, 1987. Sols analyse granulometrique des sols inorgani-
Quebec (Quebec, Que.), BNQ 2560-065. ques. Bureau de normalisation du Quebec (Quebec, Que.),
CAN/BNQ, 1986a. Sols determination de la densite relative. NQ 2501-025.
Canadian Standards Association (Ottawa, Ont.)/Bureau de NQ, 1988. Sols determination de la resistance au cisaillement
normalisation du Quebec (Quebec, Que.), CAN/BNQ 2501- non draine des sols coherents a laide du scissometre de
070. chantier. Bureau de normalisation du Quebec (Quebec,
CAN/BNQ, 1986b. Sols determination de la limite de liquidite Que.), NQ 2501-200.
a laide du penetrometre a cone suedois et de la limite de Pen, C.K., 1990. An assessment of the available methods of
plasticite. Canadian Standards Association (Ottawa, Ont.)/ analysis for estimating the elastic moduli of road pavements.
Bureau de normalisation du Quebec (Quebec, Que.), Proceedings of the 3rd international conference on bearing
CAN/BNQ 2501-092. capacity of roads and airfields, Trondheim, Norway.