Procedia Engineering

Volume 143, 2016, Pages 862–869

Advances in Transportation Geotechnics 3 . The 3rd

International Conference on Transportation Geotechnics
(ICTG 2016)

Correlating Nonlinear Parameters of Resilient

Modulus Models for Unbound Geomaterials
Mehran Mazari1, Imad Abdallah2, Jose Garibay2, and Soheil Nazarian2*
1
Savannah State University, Savannah, GA, U.S.
2
The University of Texas at El Paso, El Paso, TX, U.S.
mazarim@savannahstate.edu, emadn@utep.edu, jlgaribay@utep.edu, nazarian@utep.edu

Abstract
Characteristics of unbound aggregate pavement layers is one of the most important factors affecting the
structural performance of pavements. The resilient modulus of the unbound geomaterials has been
known as one of the most contributing factors to the mechanistic design process. The resilient modulus
material models have been evolved significantly during the past two decades. A three-parameter
nonlinear model is usually used to relate the stress state to the resilient modulus based on the
characteristics of the geomaterials. Highway agencies use different versions of the resilient modulus
models and the transformation from one model to another, requires the backcalculation of model
parameters from laboratory results. This study aims at correlating the nonlinear model parameters
between two of the resilient modulus models followed by the partial validation of the proposed process
through a number of laboratory tests. The proposed correlations were associated with some level of
uncertainty and yet were straightforward enough for practicality purposes. The performance of
prediction models, as validated with a number of laboratory results, were reasonably acceptable. These
models could then interchangeably been utilized to extract resilient modulus nonlinear model parameters
from each of the evaluated models without the need to perform the backcalculation process.

Keywords: Unbound Geomaterials, Resilient Modulus, Pavement

1 Introduction
Characteristics of unbound geomaterials play an important role on the overall performance of
pavement structures. The conventional pavement design methods were based on experimental data
collected during laboratory and large scale field evaluations. With the development of enhanced
mechanistic-empirical approaches, the mechanical properties of the geomaterials, and particularly
resilient modulus, have been implemented in the design process. Resilient modulus can be obtained
either directly, through laboratory tests, or indirectly using nonlinear prediction models, known as

*
Corresponding author

862 Selection and peer-review under responsibility of the Scientific Programme Committee of ICTG 2016
c The Authors. Published by Elsevier B.V.
doi:10.1016/j.proeng.2016.06.142
Correlating Nonlinear Parameters of Different Resilient Modulus Constitutive Models ... Mazari et al.

constitutive resilient modulus models. Such models have been under continuous development during
the past two decades. Puppala (2008) and Tutumluer (2013) contain a comprehensive synthesis of
resilient modulus models for unbound geomaterials.

Laboratory measurement of resilient modulus and estimating the model parameters are complex and
time consuming. Furthermore, the laboratory equipment is not widely available. Highway agencies use
different forms of resilient modulus material models to estimate the performance of unbound
geomaterials pavement layer. Proposing a method that can estimate the nonlinear model parameters,
would be highly desirable. Such process would save the cost of operating laboratory tests and accelerates
the process of predicting model parameters.

2 Background
Evolution of resilient modulus constitute model has been documented in a number of research
efforts. Most of these models relate the state of stress to the resilient modulus using a nonlinear
correlation with two or more regression parameters. Dunlap (1963) correlated the resilient modulus (MR)
to the confining stress and atmospheric pressure using a two-parameter power model. Drumm et al
(1990) used the deviatoric stress to predict the resilient modulus. Later, Witczak and Uzan (2000)
proposed a three-parameter nonlinear model which employs bulk stress and octahedral shear stress along
with the atmospheric pressure to estimate the resilient modulus. The National Cooperative Highway
Research Program (NCHRP) project 1-28A (Andrei et al, 2004) proposed a version of Uzan model for
use in the Mechanistic-Empirical Pavement Design Guide (MEPDG) in the US. The proposed MEPDG
model is in the form of the following equation:

k2 k3
§T · § W oct ·
MR k1 Pa ¨¨ ¸¸ ¨¨  1¸¸ (1)
© Pa ¹ © Pa ¹

where θ = bulk stress, τoct = octahedral shear stress, Pa = atmospheric pressure and k1, k2, k3 are
regression parameters.

As discussed by Mazari et al (2012) and Nazarian et al (2014), a slightly modified form of the
MEPDG model (Ooi et al, 2004) is more appropriate for estimating the responses of the modulus-based
devices. However, most highway agencies still employ the MEPDG constitutive model to estimate
nonlinear regression parameters (k1, k2, k3 in Equation 1) for unbound pavement layers. The Ooi model
is as follows:
k 2' k 3'
§T · §W ·
MR k Pa ¨¨
'
1  1¸¸ ¨¨ oct  1¸¸ (2)
© Pa ¹ © Pa ¹
where k'1, k'2, k'3 are regression parameters of the Ooi model.

The following sections include the process of converting nonlinear model parameters between the
constitutive models in Eq. 1 and 2. Development of transfer functions was followed by partial validation
of proposed process through a limited number of resilient modulus laboratory tests.

863
Correlating Nonlinear Parameters of Different Resilient Modulus Constitutive Models ... Mazari et al.

3 Methodology
The relationship between the nonlinear regression parameters of the two aforementioned models had
to be developed in a way that the proposed relationships can be used conveniently by highway agencies.
To achieve this goal, 1000 random combinations of k' parameters was generated using a discrete uniform
distribution with equal probability of outcomes and Latin Hypercube sampling method (Ye, 1998). To
simulate lab MR tests, the resilient moduli of more than a fifteen loading sequences recommended by
the American Association of State Highway and Transportation Officials (AASHTO) procedure T-307
were calculated using the modified MEPDG constitutive model (Eq. 2). The regression parameters for
the MEPDG model (k1, k2 and k3 in Eq. 1) were then backcalculated employing the nonlinear
optimization algorithm used in reducing laboratory resilient modulus test results.
A one-to-one relationship between parameters k3 and k'3 was found (i.e., k'3 = k3). Figure 1 illustrates
the correlation between parameter k2 from the MEPDG model and k'2 from modified model for the cases
evaluated in this simulation. The following global equation was found suitable to estimate k'2 from k2:

k'2 = 1.88 k2 (3)

3.5
k'2 Parameter from Equation 2

k'2 = 1.88(k2)
3.0
R² = 1.00
2.5

2.0

1.5

1.0

0.5

0.0
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8
k2 Parameter from Equation 1
Figure 1: Relationship between k2 parameter from Equation 1 and k'2 parameter in Equation 2

Figure 2 shows the relationship between simulated k1 and k'1 parameters. A global trend between
these two parameters was not found initially. Upon further analysis, it was realized that the scatter in
the data could be related to the variation in k2 parameter. Some examples of the relationships between
k1 and k'1 parameters for several discrete values of k2 parameter are presented in Figure 3. A strong
linear relationship between k1 and k'1 parameters was observed for each distinct bound of k2 parameter.

864
Correlating Nonlinear Parameters of Different Resilient Modulus Constitutive Models ... Mazari et al.

3000
k'1 Parameter from Equation 2

2500

2000

1500

1000

500

0
0 500 1000 1500 2000 2500 3000
k1 Parameter from Equation 1
Figure 2: Relationship between k1 Parameter from Equation 1 and k'1 parameter from Equation 2

3500
k'1 Parameter from Equation 2

k2=0.0
3000 K2=0.5 k'1 = 0.98(k1)
R² = 1
2500 k2=0.8
k2=1.5
2000
k2=2.0 k'1 = 2.86(k1)
k'1 = 1.42(k1)
R² = 1
1500 R² = 1

1000 k'1 = 0.34(k1)

R² = 1
500 k'1 = 5.67(k1)
R² = 1
0
0 500 1000 1500 2000 2500 3000 3500
k1 Parameter from Equation 1
Figure 3: Relationships between k1 Parameter from Equation 1 and k'1 Parameter from Equation 2
for Discrete Values of k'2

The variation in the slope of the lines observed in Figure 3 (denoted as “a”) with respect to parameter
k2 is further illustrated in Figure 4. A strong exponential correlation was observed between “a” values
and k2 parameter. Based on the results presented in Figures 3 and 4, the following relationship can be
employed to estimate k'1 parameter:

k'1 = k1 e -1.32 k2 (4)

The proposed set of equations could conveniently employed to migrate from one constitutive model
to another without the need to perform complex laboratory tests.

865
Correlating Nonlinear Parameters of Different Resilient Modulus Constitutive Models ... Mazari et al.

1.2
"a" Parameter (Slope of k1-k'1)
graph for each subgroup

1.0
specified by k2

0.8

0.6

0.4
a = e-1.32(k'2)
0.2 R² = 1.00

0.0
0.0 0.2 0.6 0.4
0.8 1.0 1.2 1.4 1.6 1.8
k2 Parameter from Equation 1
Figure 4: Relationship between slope of k1-k'1 linear regression equations (a) and k2 parameter

To partially validate the proposed models, laboratory resilient modulus test samples from three
different sources of geomaterials were prepared at different moisture contents. Each of the three
geomaterials was tested at its optimum moisture contents (OMC), at OMC±1% and OMC ±2%, in case
the OMC was less than 10%. The geomaterials with OMC greater than 10%, were tested at
OMC±10%(OMC), and OMC±20%(OMC). The AASHTO T 307-99 protocol and loading sequences
were followed for laboratory resilient modulus tests in this part of the study. Laboratory test data were
then reduced and nonlinear regression parameters were backcalculated independently using both Eq. 1
(MEPDG-recommended model) and Eq. 2 (modified model). According to the proposed process, k3
and k'3 values would be the same using both constitutive models. However, k1 and k2 values obtained
from Eq. 1 would not be the same as k'1 and k'2 values from Eq. 2. Figures 5a and 5b compares the k'
parameters, obtained directly from the backcalculation of the laboratory results, compared with those
predicted independent of the resilient modulus test results from the proposed process (Eq. 3 and 4). The
outcomes from the two processes are quite comparable given the inevitable experimental errors. This
shows that the nonlinear parameters could be transformed from one constitutive model to another, with
some level of uncertainty.
To further validate the proposed equations, representative resilient moduli (MR-Rep) values, using the
modified constitutive model with predicted k' parameters, were calculated. Since modulus is dependent
on the stress state and it is not a unique value, the NCHRP project 1-28A recommended using the
representative bulk stress and octahedral shear stress for fine grained soils as θ = 85 kpa and τoct = 21
kpa. The representative resilient moduli were then calculated for each laboratory sample at different
moisture levels. Laboratory test results were compared with the MR-Rep values using predicted k'
parameters in Figure 5c. The representative lab MR moduli from the two processes are reasonably close
for practical purposes.

USCS Atterberg Limits Moisture Density Gradation

Classification LL PI OMC1, % MDD2, kg/m3 Sand, % Fines, %
CL 27 14 10.0 1996 28 64
CH 86 53 25.9 1533 3 97
3
ML NP NP 9.4 1995 42 59
1Optimum Moisture Content, 2Maximum Dry Density, 3Non-Plastic.
Table 1: Index properties of selected geomaterials

866
Correlating Nonlinear Parameters of Different Resilient Modulus Constitutive Models ... Mazari et al.

2500
a) k'1 Parameter
Predicted k'1 parameter
from Proposed Process
2000

1500 k'1 = 1.03(k1)

R² = 0.99
1000

500

0
0 500 1000 1500 2000 2500
k'1 Parameter obtained Directly from Lab MR test
1.4
b) k'2 Parameter
Predicted k'2 parameter from

1.2
k'1 = 0.90(k1)
Proposed Process

1 R² = 0.97
0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
k'2 Parameter obtained Directly from Lab MR test
35
Predicted Representative MR
Lab Modulus from Proposed

c) Representative Lab MR Modulus

30
25
Process, ksi

20
k'1 = 1.01(k1)
15 R² = 1.00
10
5
0
0 5 10 15 20 25 30 35
Representative lab MR Modulus obtained Directly, ksi
Figure 5: Verification of Developed Process to Calculate Nonlinear Regression Parameters of Eq.
2 (k'1 and k'2) from Nonlinear Regression Parameters of Eq. 1 (k1 and k2)

4 Summary and Conclusions

Resilient modulus material models have evolved tremendously during the past decades. With the
recent introduction of the mechanistic-empirical pavement design approach, resilient modulus has
become one of the most important design parameters. Resilient modulus of a geomaterial could be

867
Correlating Nonlinear Parameters of Different Resilient Modulus Constitutive Models ... Mazari et al.

estimated from either direct methods, which involves laboratory evaluations of cylindrical samples
under repeated load in a triaxial chamber, or indirect methods, using empirical models. Since highway
agencies use different versions of the resilient modulus models, the transformation from one model to
another, would require the backcalculation of model parameters from laboratory results. Therefore,
developing a process to migrate from one constitutive model to another would be highly desirable. Such
process were investigated in this study. Two common forms of the resilient modulus models were
selected and the nonlinear regression parameters of these models were correlated using a number of
simulation studies. The proposed equations to correlate the nonlinear model parameters were then
partially validated using a series of laboratory tests on three different sources of geomaterials. The
laboratory samples were prepared at different moisture levels to further evaluate the developed process.
A strong correlation were found between the nonlinear model parameters directly backcalculated from
laboratory results and those predicted from the proposed process in this study. Furthermore, the
representative resilient moduli values calculated directly from the laboratory evaluations and indirectly
from the developed equations were reasonably close with some level of uncertainty associated with
laboratory estimations. Further validation and expansion of this study to other resilient modulus
constitutive models would be suggested in future studies.

5 Acknowledgment
This study was carried out as part of the NCHRP Project 10-84. The contents of this paper reflect
the authors’ opinions, not necessarily the policies and findings of NCHRP. The authors are grateful to
Dr. Ed Harrigan and the study panel for their help and advice throughout this study.

References
Andrei, D., Witczak, M. W., Schwartz, C. W., and Uzan, J. (2004). Harmonized resilient modulus
test method for unbound pavement materials. Transportation Research Board, 1874(1874), 29–37.
Dunlap, W. A. (1963). A Report on a Mathematical Model Describing the Deformation
Characteristics of Granular Materials. Texas A&M University, Texas Transportation Institute, College
Station, TX.
Drumm, E. C., Boateng-Poku, Y., and Pierce, T. J. (1990). Estimation of Subgrade Resilient Modulus
from Standard Tests. Journal of Geotechnical Engineering, Vol. 116, No. 5, pp. 774–789.
Mazari, M., Navarro, E., Abdallah, I., and Nazarian, S. (2014). Comparison of numerical and
experimental responses of pavement systems using various resilient modulus models. Soils and
Foundations, 54(1), 36-44.
Nazarian, S., Mazari, M., Abdallah, I., Puppala, A. J., Mohammad, L. N., and Abu-Farsakh, M. Y.
(2014). Modulus-Based Construction Specification for Compaction of Earthwork and Unbound
Aggregate. NCHRP Project 10-84, El Paso, TX.
Ooi, P. S. K., Archilla, A. R., and Sandefur, K. G. (2004), Resilient Modulus Models for Compactive
Cohesive Soils. Transportation Research Record: Journal of the Transportation Research Board, No.
1874, pp. 115-124.
Puppala, A.J. (2008), Estimating Stiffness of Subgrade and Unbound Materials for Pavement Design,
NCHRP Synthesis 382, National Cooperative Highway Research Program, Transportation Research
Board, Washington, D.C.
Tutumluer, E. (2013). Practices for Unbound Aggregate Pavement Layers. NCHRP Synthesis 445,
National Cooperative Highway Research Program, Transportation Research Board, Washington, D.C.

868
Correlating Nonlinear Parameters of Different Resilient Modulus Constitutive Models ... Mazari et al.

Witzack, M. W., and Uzan, J. (2004). Harmonized Test Method for Laboratory Determination of
Resilient Modulus for Flexible Pavement Design. NCHRP Project 1-28A, TRB, National Research
Council, Washington, D.C.
Ye, K. Q. (1998). Orthogonal column Latin hypercubes and their application in computer
experiments. Journal of the American Statistical Association, 93(444), 1430-1439.

869