Transportation Geotechnics 23 (2020) 100326

Contents lists available at ScienceDirect

Transportation Geotechnics
journal homepage: [Link]/locate/trgeo

Mechanistic-empirical permanent deformation models: Laboratory testing, T

modelling and ranking
Ana Ramosa, António Gomes Correiaa, , Buddhima Indraratnab, Trung Ngob, Rui Calçadac,
⁎

Pedro Alves Costac

a
University of Minho – School of Engineering, Campus de Azurém, 4800-058 Guimarães, Portugal
b
University of Wollongong Australia – School of Civil, Mining and Environmental Engineering, Northfields Ave Wollongong, Wollongong, Australia
c
Faculty of Engineering of University of Porto, Rua Dr. Roberto Frias, 4200-465 Porto Portugal

ARTICLE INFO ABSTRACT

Keywords: Geomaterials exhibit elastoplastic behaviour during dynamic and repeated loading conditions. These loads are
Mechanistic-empirical permanent deformation induced by the passage of a train or vehicle which then generates recoverable (resilient) deformation and/or
models permanent (plastic) deformation. Modelling this behaviour is still a challenge for geotechnical engineers as it
Permanent deformation implies the understanding of the complex deformation mechanism and application of advanced constitutive
Laboratory tests
models. This paper reviews on the major causes of permanent deformation and the factors that influence the
Ranking
long-term performance of materials. It will also present the fundamental concepts of permanent deformation as
well as the models and approaches used to characterise this behaviour, including: elastoplastic models, shake-
down theory and mechanistic-empirical permanent deformation models. This paper will focus on the mechan-
istic-empirical approach and highlight the evolution of the models, and the main similarities and differences
between them. A comparison between several empirical models as well as the materials used to develop the
models is also discussed. These materials are compared by considering the reference conditions on the type of
material and its physical state. This approach allows for an understanding of which properties can influence the
performance of railway subgrade and pavement structures, as well as the main variables used to characterise this
particular behaviour. An innovative ranking of geomaterials that relate to the expected permanent deformation
and classification (UIC and ASTM) of soil is also discussed because it can be used as an important tool for the
design process.

Introduction structures avoid a mediocre performance [6,54,59]. This is why the

permanent deformation of geomaterials should be included in the de-
Pavements and railway structures are designed for a certain lifespan sign since otherwise, it can lead to higher annual rehabilitation costs
and they usually require reconstruction and/or major maintenance [59].
work because the materials may experience fatigue, cracking or be Over the past decades, researchers have been so concerned about
under an excessive amount of cumulative permanent deformation (or permanent deformation and they continually search for the most ac-
rutting in the case of pavements) due to the passage of traffic. Plastic or curate methods and models that will measure and predict these values
permanent deformation usually occurs in the geomaterials (base, sub- [54,48,72,30], among others. To achieve these objectives, laboratory
base or subgrade soils) which are responsible for the surface rutting and investigations using cyclic triaxial tests, simple and cyclic shear tests,
that can lead to significant passenger discomfort [60]. This is why resonant column and hollow cylinder tests, among others have been
knowing and understanding the deformation and failure mechanisms of carried out. These tests were often used to determine shear stress–strain
geomaterials under dynamic and cyclic loading is so important when behaviour, resilient modulus (Mr) of subgrade geomaterials considering
designing and planning the maintenance of pavements and railway reversible and irreversible deformation under cyclic loads [32,24]. The
structures [49]. Indeed, an accurate estimation or prediction of the reversible (elastic) is usually described by non-linear elastic models, but
amount of cumulative settlement will help pavements and /or railway since permanent deformation is more complex, it depends on the

Corresponding author.
⁎

E-mail addresses: id6629@[Link] (A. Ramos), agc@[Link] (A. Gomes Correia), indra@[Link] (B. Indraratna),
trung@[Link] (T. Ngo), ruiabc@[Link] (R. Calçada), pacosta@[Link] (P.A. Costa).

[Link]
Received 16 October 2019; Received in revised form 24 January 2020; Accepted 26 January 2020
Available online 28 January 2020
2214-3912/ © 2020 Elsevier Ltd. All rights reserved.
A. Ramos, et al. Transportation Geotechnics 23 (2020) 100326

Notation pini mean stress in the initial state of the material

qini deviator stress in the initial state of the material
UIC International Union of Railways σ1 orσd cyclic vertical stress or cyclic deviator stress
ASTM American Society for Testing and Materials σ3 or σ3 Horizontal/confining stress
SS single-stage σ1/σ1:f failure ratio of the major principal stresses
MS multi-stage σoct octahedral normal stress
RLT repeated load tests τoct octahedral shear stresses
Mr resilient modulus m and s defined by the Mohr-Coulomb yielding criterion
N load cycles q = s + mp; m is the slope
εp permanent strains WL liquid limit
p mean stress WP plastic limit
q deviatoric stress IP index plasticity
qf deviatoric stress at failure Cu coefficient of uniformity
pam amplitude of the mean stress for train loadings Cc coefficient of gradation
qam amplitude of the deviator stress for train loadings

accumulation of N loading cycles. In order to select the best testing approach and the most suitable
One of the main objectives of this paper is to review on the main model, it is important to understand the conditions needed for its de-
causes of permanent deformation on pavements and railway structures velopment such as the properties of the materials tested, degree of
and the factors that can increase this phenomenon. It is noted that while compaction, moisture content, etc., as well as the main variables/fac-
the materials should be able to resist permanent deformation, this re- tors that can influence the response of the material. This paper reviews
sistance will depend on the number of load cycles and stress levels [46], on the existing methods used to estimate the irreversible deformation of
the thickness of the layer, and the granulometry of the material. This geomaterials, followed by a parametric study that includes comparisons
also includes other external factors such as the physical state of the soil, among some selected models on different materials with different
which is often difficult to control because it depends on other en- classifications (UIC and ASTM), properties, granulometry, and physical
vironmental aspects such as the moisture content, and degree of sa- states. This comparison allows to estimate the permanent deformation
turation, etc. and to rank materials according to the predicted deformation data and
The behaviour of geomaterials under cyclic loads can be char- soil classification, which is a helpful tool in the design of the pavement
acterised by either using complex elastoplastic models (recoverable and and railway structures. It is also noted that this ranking should be in-
permanent deformation are both considered) or by shakedown theory terpreted as a reference value because it depends on several properties
and mechanistic-empirical models [37]). The elastoplastic models, and soil conditions.
however, despite their ability to accurately predict permanent de-
formation (the loading history is considered because the equation is
solved based on incremental steps), they are difficult to implement, Causes of permanent deformation
time consuming, and complex [50]. Most of these models only consider
a low number of load cycles, which is not in accordance with the in situ Permanent deformation occurs in pavements and railway lines due
conditions where the number of loads is up to million cycles. Indeed, to repeated traffic loading; and if the volume of traffic is high enough it
these models are very demanding computationally because they require may lead to permanent deformation and/or structural failure.
the simulation of repeated load applications in pavements/railway Therefore, it is crucial to identify the main causes of permanent de-
structures. The development of formulations based on cyclic con- formation and adopt appropriate measures, even during the design
stitutive laws may be expressed through conventional concepts such as process. Furthermore, transition zones are also important areas due to
the yield condition, hardening and flow rules. The main problem with the possible development of differential settlement due to differences in
the numerical implementation is that the increment of permanent de- stiffness between two contiguous structures such as between a ballasted
formation per cycle becomes very small quickly, and this leads to and slab track, or a bridge and a plain track (Fig. 1). In fact, the de-
problems with the computational accuracy of the results [1]. The focus gradation in these zones often accelerates because the differential set-
of this work will be on mechanistic-empirical models. These models are tlement increases the dynamic effects [40].
based on extensive laboratory testing results, so they can correctly si- Another important cause of permanent deformation is related to the
mulate the response of materials; they are easy to implement, and they complex stress conditions as geomaterials are subjected to vertical, hor-
depend on fewer parameters than conventional elastoplastic models. izontal and shear stresses during the passage of vehicles as well as the effects
of moving loads; all of which implies the rotation of principal stresses.

Fig. 1. Example of a rail track at a transition zone.

2
A. Ramos, et al. Transportation Geotechnics 23 (2020) 100326

Flexible road pavements often consist of an upper layer of asphalt over the number of load cycles (N) and stress levels. While these factors can be
unbound granular layers (base, subbase), which have been compacted over divided into load-related factors and material properties, the load factors
a proper soil subgrade. These unbound granular layers provide a structural will include applied stress levels, number of load applications, the strength
role, but in some cases when subgrade has an adequate bearing capacity, of material, as well as the loading history and the effect of principal stress
rutting still occurs in the granular layers causing fatigue cracking in the rotation. However, other factors can also influence the plastic strain:
bituminous layers. This means the bituminous layers and the subgrade will moisture content (degree of saturation), matric suction, fine content, density
lead to permanent deformation. In rigid pavements, only the subgrade leads (degree of compaction), aggregate type, particle size distribution (grada-
to the development of permanent deformation. tion), and the amount and type of fines (plastic or non-plastic) [45,82].
Post settlement in railway structures is often caused by the self- Gidel et al. [30] found that for unbound granular materials, the miner-
weight of the embankment and traffic loading. Permanent deformation alogical nature of material (including aggregate mineralogy and particle
includes the settlement of the roadbed beneath the track and the sub- morphology) also influences permanent deformation [22]. This factor has
grade. Several studies show that the main influencing factors are the important effects on particle shape, the plastic and nonplastic nature of fines
level of stress and stress history; in fact, the experimental results and content, and sensitivity to water and surface roughness [61]. However, the
field measurements show that the dynamic and cyclic loads from train stress level and number of load cycles emerge as the most important factors.
passage make a significant contribution. It is noted that although both In fact those models that only consider the value N should not be used to
structures are affected by permanent deformation, pavements and predict permanent deformation because they are too simple and lacks ac-
railway structures have different dynamic effects on the structure where curacy.
the dynamic effects are more severe on railway structures. In specific cases of pavements and railway structures, the stress con-
Permanent deformation has a significant effect on the performance of ditions are important because the track foundations are subjected to
structures because it leads to an increase in maintenance operations and complex vertical, horizontal and shear stresses. Furthermore, railways and
costs, and reduces ride quality. Despite this is long-term behaviour, it must pavements present other complex problems related to the effects that
be predicted during the design stage because although the accumulation is moving loads have on track foundation; for instance, the rotation of prin-
very small during each cycle, it may still lead to the ultimate collapse of the cipal stresses affects plastic strain. Chan [12] showed an increase in per-
structure (excessive rutting) due to the accumulation of millions of cycles. manent deformation when the rotation of principal stresses was included.
Figs. 2 and 3 show that the permanent deformation can be defined and These tests took place in a hollow cylindrical apparatus which allows for
characterised by the accumulation of small increments of deformation simulation in the laboratory and similar conditions to be verified in situ.
during N loading cycles. Several studies show that permanent strains de-
pend directly on the mean and deviator (p and q, respectively), levels of
stress [47]. The mean stress is dependent on the sum of the principal stresses Laboratory testing
while the deviator stress is dependent on the sum of squares of the differ-
ences of the principal stresses: There are a number of laboratory tests currently used to evaluate
the permanent deformation of geomaterials; they attempt to reproduce
1 + 2 + 3 in situ stress conditions in pavements and railway structures. The cyclic
p=
3 (1) triaxial test is the most widely used to study of geomaterials subjected
1 to cyclic loads. However, in these tests, the principal stresses are always
horizontal or vertical, which may not always correspond to in situ
q= × ( 2 2 2
1 2) +( 2 3) +( 3 1)
2 (2)
conditions where the materials are subjected to moving loads and ro-
tations of principal stresses. The cyclic vertical stress, designated as σd,
Factors influencing the permanent deformation is applied, and horizontal stress, designated as the confinement stress -
σ3 is also applied. This test begins when deviator stress is applied, and
Permanent deformation is a complex process that directly depends on then the applied vertical progressively increases until it reaches an

Fig. 2. Illustration of the accumulation of permanent deformation under cyclic loads (after two cycles).

3
A. Ramos, et al. Transportation Geotechnics 23 (2020) 100326

Fig. 3. Accumulation of permanent strain (adapted from Erlingsson et al. [29]

allowable displacement of the apparatus. [84]. Studies that relate the two parameters (Mr and εp) have shown that
Another test used to evaluate permanent deformation is a cyclic for subgrade soils, Mr has a significant effect on the εp or pavement rutting
torsional test, which also includes a hollow cylinder test. This particular since soils with a higher Mr present less permanent strain [55,65]. How-
test enables the magnitude and the direction of principal stresses to be ever, mixed soils such as silty sand or sandy silts still show significant
controlled. In fact, this test device has been developed to study per- rutting despite having higher resilient characteristics [72,60].
manent deformation and resilient behaviour by considering the rotation Existing permanent deformation models can be divided into two main
of principal stresses, and then applying the laboratory results to in situ categories: rutting models based purely on mechanics, and mechanistic-
conditions. In this case, the specimen is cylindrical and hollow. The empirical models. Rutting models are based on elastoplastic theory
main limitations of this test are the dimensions of the sample and the [25,26,74,73,13,15,16]. While they can consider how the stress levels and
stress path can only be simulated on the symmetrical plane of the track. stress paths can affect permanent deformation, they are very complex and
However, the results show that the permanent deformations are often difficult to use because: (i) the increment of permanent strain per cycle is
higher in this test than the convention cyclic triaxial tests due to the very small; and (ii) the response of the track in each cycle is a challenge for
simulation of the rotation of principal stresses. cyclic constitutive models [16]and numerical implementation [1]due to
Cyclic shear tests can also be used to study the dynamic behaviour the extensive calculation time and cumulative errors.
of soil. In this test the sample is cylindrical and a horizontal shear force Mechanistic-empirical models often describe a relationship between
is applied to the bases of the samples. This test can also be used to the number of load cycles and the accumulated permanent deformation.
evaluate stability under seismic events, to quantify the degradation of They are very simple to use, the numerical results can be obtained very
shear stress in cohesive soils under cyclic loads, and to evaluate the quickly, and they also predict permanent deformations very well.
liquefaction parameters of non-cohesive soils under cyclic loading. However, one of the particularities of these models is that their deri-
There are also other approaches for studying the permanent de- vation from triaxial tests; which means the conventional and well-
formation of geomaterials; for example, while expensive, physical known heart-shaped stress path (depicted in Fig. 4) induced by the
models (1-to-1 prototype model test) can be used to evaluate the per- passage of trains cannot be reproduced in this type of laboratory test.
formance of geomaterials and the stability of structures subjected to Mechanistic-empirical models can be divided into single-stage
dynamic and repeated loading. These models will provide actual field models and multi-stage models. A single-stage implies that the re-
data to help understand the behaviour of railways and pavements under petitive load tests are carried out at one stress level in one test; in this
moving loads, and enable the study of permanent deformation and the instance, multiple specimens are tested at different stress levels. Multi-
stresses induced in soil through proper measurement devices. stage models can test multiple levels of stress in one test on one spe-
While small scale models of railways were introduced in the past [53,2], cimen. This approach enables the effects that the stress level and stress
several models focusing on the development of permanent deformation of history have on permanent deformation to be considered [34].
railways under cyclic loadings or moving loads at low speed [7]are currently
available. However, the field measurements and numerically calibrated re-
Shakedown theory
sults show that increased train speed has a huge influence on the long-term
performance of a structure (i.e., higher dynamic stresses and higher per-
Shakedown theory is another approach used to characterise load-
manent deformation). A test facility should be able to simulate the actual
deformation responses of geomaterials. This theory was developed in-
speed of a moving train under very large number of load cycles. This means
itially based on the behaviour of metals subjected to repeated loading.
that full-scale models are better because the measured results from a test can
be considered directly in the track design and maintenance operations as
this approach reduces any uncertainties in the measurements [7].

Modelling approaches

Permanent deformation can be predicted either by numerical si-

mulations using elastoplastic models utilising the shakedown theory or
mechanistic-empirical deformation models based on laboratory tests
such as cyclic triaxial tests or hollow cylinder apparatus. In addition,
the accumulated permanent deformation can be measured by the re-
peated load triaxial (RLT) tests developed in the laboratory.
Mechanistic-empirical permanent deformation models have become
much more complex due to the inclusion of other variables, apart from the
number of load cycles and stresses. More recently, a study to correlate the
resilient modulus (Mr) with the permanent strains (εp) has been developed Fig. 4. Stress conditions due to a single moving wheel load [33]

4
A. Ramos, et al. Transportation Geotechnics 23 (2020) 100326

Werkmeister et al. [79] further developed to apply for unbound gran- assessment of the load that is compatible with shakedown state. The lower
ular materials. The cyclic loads may not lead to instantaneous collapse bound theorem states that an elastic perfectly plastic material will shake-
of the structure since they can induce plastic strain in the material in down if it is possible to find a time-independent, self-equilibrated, residual
every load cycle (N). Indeed, if the load level is lower than a critical stress field that combined with the cyclic elastic stresses and possible static
limit, the material will show permanent deformation in the first load elastic stresses gives rise to a stress field that no-time and nowhere violates
cycles. However, after a certain number of N, the material will respond the yielding criteria [3].The shakedown will occur if the following condition
elastically to the subsequent load cycles. This phenomenon is defined as is experienced (any time and any location):
shakedown and this critical limit is called shakedown limit. e o r
f( ij + ij + ij 0 (3)
Here, the material is divided into three categories (by considering its
stress dependency): range A, range B, and range C, as shown in Fig. 5. where f represents the yielding criterion, λ is a load factor of stress field, ijo
Werkmeister [77] defined the shakedown limits based on the re- is the rest stress state, ije are the elastic stresses induced by the cyclic loads
peated load triaxial test (RLT); these values are also defined by and the residual stress field is given by ijr . The maximum value of λ, cor-
EN13286-7 (2004a): responds to the lower value of the shakedown multiplier. The kinematic
shakedown theorem states that if any kinematic acceptable mechanism of
• Range A: ( 5000 3000 )
< 0.045 × 10 3 plastic deformation can be found, the shakedown phenomenon can not
• Range B: 0.045 × 10
p p
3
< ( p5000 3000 ) < 0.4 × 10 3 occur since the structure will fail due to fatigue [3].
• Range C: (
p
p
5000
p
3000 ) > 0.4 × 10 3 Having realised the importance of elastoplastic models and shake-
down theory as approaches to predict permanent deformation, the
These limits represent the deformation that accumulates between mechanistic-empirical models are also an essential approach and will be
the 3000th and 5000th number of load cycles. presented in the next section. These models are simple and they present
Fig. 5 shows that Range A is designated as plastic shakedown where an elegant formulation with good results, particularly when compared
the permanent deformation stabilises after a finite number of load cy- to the laboratory tests and in situ measurements.
cles and the material becomes completely resilient. In fact, after a
certain point, the cyclic stress does not cause any damage to the Mechanistic-empirical permanent deformation models
structure, which means that the permanent deformation does not in-
crease and the failure does not occur. Range B is designated as the in- The mechanistic-empirical model is often derived from laboratory
termediate-range where accumulated deformation increases without test results such as the triaxial cyclic tests, direct shear tests or large-
complete stabilisation. In this case, permanent deformation develops at scale cubical tests. There have been a number of predictive models used
a very slow rate and is almost linear. Range B occurs when the repeated to study permanent deformation range from purely empirical to me-
load cycles overcome plastic shakedown and failure can occur after a chanistic and plasticity theory-based models. However, some of these
large number of load cycles. Range C is defined as incremental collapse models are only applicable to specific stress states or testing conditions
because the permanent strain accumulated per cycle increases until the so they have never been evaluated for a wide range of stress states, or
failure occurs. This is caused by shear failure associated with the re- types of materials and their physical conditions [82].
orientation and rearrangement of particles, breakage, and the slip and It is noted that most of the developed models considered total stress
loss of friction between the particles [68]. Indeed, failure can occur conditions, while ignoring the importance of suction. With regards to
after a low number of loading cycles. This type of behaviour should the resilient modulus (Mr), there have been several studies about the
never be accepted, unlike materials in the range A or even range B. It is influence that suction has on the Mr even though this influence is not
reported in recent studies that the shakedown limits were used for the widely used in the study of plastic strains [23]. Salour and Erlingsson
pavement analysis [62,64]and in railway structures [3]. [67] used a triaxial testing system to control the pore-water pressures of
The shakedown analysis is used to find the shakedown limit of a specimen, so the tests for permanent deformation took place with the
structure under cyclic load. In the case of railway structure, the sha- control matric suction of the soil samples, and an effective stress ap-
kedown is used to predict whether the settlement will keep increasing proach has been used for modelling.
or reach a stable status. In past years, some laboratory studies were The mechanistic-empirical permanent deformation models can be
performed [78,8]on different types of materials (granular and soils) in divided by considering the materials used in the tests, the approach
order to evaluate the possibility of shakedown under traffic loads [51]. used during the tests, and the complexity (variables included in the
Indeed, the shakedown approach has become very popular in pavement model), as illustrated in Fig. 6 and described in the following sections.
engineering, where some progress had been reported [21,14,17,18,8].
Thus, the main concern consists in to prevent excessive permanent Materials
strain in a certain structure and guarantee that the loading level is
below the elastic shakedown limit. Indeed, the assessment of this limit Empirical permanent deformation models can be categorized ac-
is not easy since it demands a significant computational effort mainly in cording to materials tested as cohesive and granular materials. In the
3D modeling since the loading time-history needs to be considered. Annex, Table A1 and Table A2 present summaries of the empirical
Considering the classical limit analysis theory, two main theorems were
formulated to define the limits of shakedown load; the lower bound the-
orem (conservative solution) and the upper bound theorem (unconservative
solution). The application of the lower bound theorem demands an accurate
evaluation of the elastic response of the track’s foundation due to the traffic.
A numerical model can be used to assess the elastic dynamic response of the
system [3]. During the shakedown phenomenon, the structure is subjected
to a load that is lower than this limit but higher than the elastic limit and,
after the accumulation of permanent deformation in the beginning (finite
number of cycles), the material starts to show elastic response and after this
point, there is no accumulation of permanent deformation. This behavior is
explained by the total stress field that is given by the sum of the residual
stress field with origin in the previous cycles. These conclusions were also
obtained by Wang and Yu [75]. The key point of this problem is the Fig. 5. Illustration of shakedown theory [29]

5
A. Ramos, et al. Transportation Geotechnics 23 (2020) 100326

models that characterise the permanent deformation of cohesive ma- granulometry (poor or well graded) can also influence the response of
terials, fine-grained soils–clays and grained soils–silts, respectively. In granular materials, as well as their physical state [31]. However, in most
addition, Table A3 and Table A4 present key features of empirical empirical models these variables are not directly considered.
models that can be used to obtain the irreversible deformation of
granular materials classified as sands and sandy gravels, respectively. Approach

Cohesive materials Based on approach and procedures for a cyclic load test, mechanistic-
The plastic strain that accumulates in cohesive materials is a result empirical models can be categorised as either single-stage (SS) or multi-
of the plastic deformation of shear strain, the accumulation of strain stage (MS) models. Single-stage models mean that repeated load test (RLT)
due to compaction, consolidation, and associated residual accumulation are performed at one stress level in one test, while multi-stage models
of excess water pressure. Models for cohesive materials should consider mean that RLT tests are performed at multiple stress levels in one test on
the stress state, the type of soil, and the physical state of the materials, one specimen. This approach is a relatively recent one which considers
and variables related to the moisture content and dry density. These how the stress level and the stress history will affect permanent de-
types of materials are also influenced by environmental conditions and formation [28]. Actually, a multi-stage RLT better represents reality and
traffic loading, so regardless of the physical state of the material, its the real conditions of soils in the field that are subjected to cyclic loads.
influence is considered in most models, and indirectly in the material With the MS approach, the influence of different stress paths with different
parameters/constants, as summarised in the Table A1 and Table A2. magnitudes (representing the influence of the effects of stress history) can
The shakedown behaviour of cohesive soils is characterised by the be analysed. This approach is referred to in EN13286-7 [27] and it allows
stress–strain loop, and when the load increases, the progressive transition of for the application of a certain number of consecutive stress paths to the
material from plastic shakedown through plastic creep to incremental col- same specimen, thus reducing the time spent in the laboratory, but it also
lapse can be identified [83]. Yang et al. [84] indicated that it is not possible increases the complexity compared to the single-stage RLT. Gidel et al.
to identify the pattern for the cohesive materials shown in Fig. 5. Under [30] introduced this concept into their work, as did Erlingsson and
lower levels of stress, the subgrade (of a pavement or railway structure) will Rahman [28] noting that their work was based on the time-hardening
reach an equilibrium strain state after a certain number of load cycles. In concept introduced earlier by Lytton et al. [52] and Gidel et al. [30]. In
these cases, plastic strain accumulates very slowly as the number of load this method the permanent deformation formulation is modified [28]:
cycles increases. As expected, the rutting of pavements or even the failure of
p = f1 (N ) f2 (p , q, r) (4)
subgrade can be prevented by controlling the accumulation of excessive
plastic strain by keeping repetitive load levels below the critical stress. Tests to
results have shown that subgrade soils can easily accumulate excessive
= f1 (N Ni + Nieq) f2 (pi , qi , ri ) (5)
plastic strain at high water content, but this will vary depending on the p 1

environmental conditions [83]. where, the subscript i refers to the ith stress path and Nieq is determined by
considering the following expression:
Granular materials
Nieq = f3 ( pi 1, pi , qi , ri ) (6)
There have been substantial studies on the use of permanent de-
formation models for granular materials because of their common use in where, pi 1 is the accumulated permanent strain at the end of the (i-1)th
pavement/railway structures [41]. The study of granular materials sub- stress path.
jected to repetitive loads is fostered by the gradual accumulation of mul- Despite the reduction in time, most of the models presented in the
tiple increments of plastic deformation (Fig. 3), which could lead to pa- bibliography are developed under the single-stage RLT like the
vement failure due to excessive rutting [47]. In order to justify the use of Korkiala-Tanttu [44] model.
these materials in pavements, some researchers tried to correlate repetitive
loading with simple static loading tests [48], but there was no consensus Complexity
with this approach because granular materials respond differently under
static and cyclic loads [69]. Other studies also tried to correlate resilient Empirical permanent deformation models have been changing over
and plastic deformation through a mathematical expression. time with an increase in their complexity (variables included in the
With regard to resilient behaviour, researchers choose different ap-
proaches. Jouve et al. [42] proposed decomposing axial and horizontal
stresses and strains into volumetric and shear components, even though
this expression depends on the permanent volumetric and shear strain
(N > 100) with the mean and deviatoric stresses. Other studies have
expressed permanent strain as a function of the number of load cycles
[6,69,80,43,57,58], and some researchers related permanent strain with
the stress level and stress ratio [9,6,56]. Lekarp et al. [47] found a de-
pendency between permanent strain, and the maximum stress ratio and
length of the stress path in p-q space. In fact, applied stress is one of the
primary factors that can influence the permanent deformation of ag-
gregates [82]. To simulate the correct response of granular materials
subjected to cyclic loads (similar to in situ conditions) in a laboratory,
variable confining pressure triaxial tests are a practical approach because
the effect of horizontal and vertical stresses can be combined.
The behaviour of granular materials under cyclic loads is complex due
to the gradual accumulation of permanent strain with the number of load
cycles. When a cyclic load is applied to a material there is a gradual ac-
cumulation of plastic deformation, a reduction in the number of voids, and
an increase in stiffness [28,5]. As with cohesive soils, the shakedown
theory can be used to classify the response of the material in terms of
permanent deformation. Furthermore, physical characteristics such as Fig. 6. Division of the mechanistic-empirical permanent deformation models.

6
A. Ramos, et al. Transportation Geotechnics 23 (2020) 100326

model). Most models establish a relationship between permanent de- influence of initial stress indirectly and explicitly [11,19,50,76] that
formation with the number of load cycles and the levels of stress. Some can explain the apparent decrease of permanent deformation as the
initial studies on resilient and permanent strains only considered the initial mean stress increases. Put simply, a higher initial stress leads to a
number of load cycles [6]. Despite their simplicity, these models have greater distance to the failure criterion which then leads to a lower
some historical significance and enable us to understand why stresses value of permanent deformation. Some other recent models based on
are an important factor when determining permanent deformation. In the power-law, εp = A·Nb include, among other factors, the octahedral
fact, some of these models did consider the stress conditions indirectly normal and shear stresses (σoct and τoct) instead of the p and q in the
in their constants, as presented in Table 1. Some studies showed the formulation of the model [59,60,10]. There are more complex models
importance of the stress state and its influence was considered in the that also include the influence of the physical state of the material, as
permanent deformation models, as shown in Table 2. It is noted that for example, the model by Xiao et al. [82], but this model depends on
most of the empirical permanent deformation models the stress levels several parameters that can be very difficult to obtain.
are defined through the deviatoric (q) and mean (p) stresses. To simulate and predict permanent deformation, the most suitable
More recently developed permanent deformation models include sev- model for each analysis/situation must be chosen, therefore the model
eral improvements over the former ones. Indeed, the first model only must be selected according to the material to be tested and its physical
depended on the number of load cycles (N), whereas later models de- condition. The model must then be evaluated according to its com-
pended on the stress levels and some included the influence of stress in- plexity in terms of modelling to the number of variables required by the
directly (through their constants); these models are still simple as those formulation, and to its efficiency and accuracy with regards to expected
developed by Monismith et al. [54], Sweere [69] and Huurman [38]. results. A simpler model could easily represent in-situ conditions rather
Most models depend directly on the mean (p) and deviatoric (q) than complex models that depend on several variables that may be
stresses and evaluate the length and slope of the stress path through the needed to carry out certain laboratory tests.
relationship established between q and p, as does the model developed
by Hyde [39], Lekarp and Dawson [46] and Rahman and Erlingsson
[66]. This model is dependent on the factor Sf: A comparison of permanent deformation models – parametric
study
p (N ) = N bSf Sf

This section attempts to compare different permanent deformation

S =
() q
pa models available for different types of soils considering the model devel-
oped by Chen et al. [19]. Whereas this comparison depends on the soil
()
f
p
pa (7) classification, the results go beyond the soil type. Indeed, while two soils
can be integrated into the same classification (UIC or ASTM), the labora-
Nevertheless, these models have considered the influence of stress
tory conditions may differ greatly and therefore lead to different results.
dependency by including the stress ratio; for example, Huurman [38]
It is noted that Chen et al. [19]’s model considers the stress level
included the effect of the stress ratio on one of his model constants by
(the model includes the amplitude of diagram p-q), the number of load
considering the failure ratio of the major principal stresses σ1/σ1:f. This
cycles, the initial stress state, and it quantifies the proximity of the peak
formulation means the deformations are larger when the failure ratio is
stress point (pmax, qmax) to the yielding criterion. This particular model,
close to failure. The model defines the ratio as the relationship between
therefore, includes several important conditions in its analysis of per-
deviatoric stress and deviatoric stress at failure (q/qf) because the de-
manent deformation.
viatoric stress is supposed to be the most dominating stress component
For all the experimental data, calibration was performed to find the
for permanent strains [49,11,44,20,82,85,35]. The model developed by
best fit for the experimental data through the parameters εp;o, α and B
Gidel et al. [30] and further extended by Chen et al. [19] is more
(correspond to material properties), presented in the following equation:
complex because it depends on the failure criterion by including the
strength parameters m and s of the Mohr-Coulomb yielding criterion.
Subsequently, there have been several models including the

Table 1
Empirical models dependent on the number of load cycles.
Authors Permanent Strains Parameters

Barksdale [6] p
1 = a + blog (N )
- a and b are constants
Khedr [43] 1
p - b is a material parameter
=A N b - A is a material and stress–strain parameter given as a function of shear stress ratio and resilient modulus
N
Paute (1988) p N p - A0 – parameter function of the stress level,
1 = A0 + 1 (100)
N +D - D is a regression parameter
- 1p (100) – permanent deformation after the first 100 cycles
Tseng and Lytton - εp is the permanent strain
(1989b) p = p N - 0p is the maximum permanent strain
0 e
- N is the number of load cycles
- ρ is the scale factor
- β is the shape factor
Sweere [69] log ( 1p (N )) = a + blog (N ) - a,b – material parameters
* for each stress level, separate permanent strain parameters a and b need to be determined from the results of
cyclic load triaxial test**The Sweeres’s formula is valid for granular materials [30]
Hornych (1993) - A and B
( )
p N B p
1 (N ) =A 1
100
+ 1 (100) - 1p (100) - permanent deformation after the first 100 cycles
Vuong (1994) - a, b and c
1
p
= r
1 ( )N
a
b
c
- 1r is the resilient axial strain
Wolff and Visser [80] 1
p
= (cN + a)(1 e bN ) - a, b and c are regression parameters
Huurman [38] - A, B, C, D are parameters function of stress level
( )
p N B N
1 (N ) =A + C eD 1000 1
1000

7
A. Ramos, et al. Transportation Geotechnics 23 (2020) 100326

Table 2
Empirical permanent deformation models dependent on the number of load cycles and stress levels.
Authors Permanent Strains Parameters

Barksdale [6] p q a . 3b - a. 3b is a relationship defining the initial tangent modulus as a function of confining pressure (σ3);
1 = a and b are constants
Rf q (1 sin )
1
2(C cos + 3sin ) - Rf is the ratio of the applied deviator stress at failure qf
- ϕ is the friction angle and C is the cohesion
Hyde [39] p
=a
q - a is a constant;
1
3 - σ3 is the confining stress;
- q is deviator stress.
Shenton (1974) - K and α
1
p
=K ( ) qmax
3 - qmax is the maximum deviatoric stress applied
Pappin [56] p q0
2.8 - fn(N) – depends on the number of cycles (shape factor);
1 = fn (N ) L - q0 is the modified deviator stress ( 2 3 q)
p0
- p0 is the modified mean normal stress ( 3 p )
- L = p2 + q2 (length of stress path)
Lentz and Baladi [48] - ε0,95S is the axial strain at 95% of the deviatoric stress at failure
(1 )
p q 0.15 n (q S )
= 0.95S ln + ln (N)
1 S 1 m (q S ) - m is the slope of the failure line
- S is the deviatoric stress at failure
Tseng and Lytton p p ( N) - εp is the permanent strain of the granular material;
= 0 e
[70–71] 1
- 0p is the maximum permanent strain;
- N isthenumberofloadcycles;
- ρ is the scale factor;
- β is the shape factor.
Paute et al. [58] q - 1p is the permanent axial strain after the first 100 cycles
p (p + p )
1 = f (N ) A = f (N ) - b – regression parameter
q
b m
(p + p )
- p* is a stress parameter defined by the intersection of the static failure line and p-axis in p-q space
- m is the slope of the failure line in p-q space
- f(N) function of the number of cycles N
Nishi (1994) p qa - a and b
=k
1, ult pb - 1,pult is the ultimate permanent axial strain
Lekarp and Dawson [46] p - a and b
()
b
1 (Nref ) q
=a - Nref
(L p0 ) p
max
- L = q2 + p2 ,
- p0 = 100kPa (reference mean stress)

p
1 (N )
12 kPa, whereas with the Chen et al. model, the specimens are saturated
2 2
a and the boundary is the drainage condition. With the Rahman and
pam + qam 1
= p0
1 [1 e BN ] Erlingsson [66] model, the influence that the moisture content has on
pa m 1+ ( pini
pam )+ s
pam
(qini + qam)
pam (8)
permanent deformation, and the degree of saturation, is analysed but
this analysis only considers the optimum moisture content for each type
of soil. Confining pressure, σ3 is tested as 27.6, 41.4 and 55.2 kPa, and
where, pam and qam are the amplitudes of the mean stress and deviator
cyclic deviator stress varies according to the stress path. Thus, the de-
stress for train loadings, m and s are defined by the yielding criterion
velopment of these models includes laboratory conditions such as soils
q = s + mp; and pini and qini are the mean and deviator stress in the initial
under different state conditions and different stress conditions.
state of the material.
Whereas Huurman [38] considers seven poorly graded sands and
These selected materials are representative of different types of mate-
one well-graded sand (designated as crusher), only C. Bruynweg and
rials (silt and sand), the percentage of fines, and the granulometry (poor and
Crusher corresponds to the poor and well-graded sands are analysed
well graded sands), as depicted in Table 3. The models are calibrated
(Table 4). In addition, the C. Bruynweg was selected as the stress ratio is
considering the same stress paths for all materials during the calibration
similar to the one carried out by Chen et al. [19]. The remaining sands
process. The stress path described by Chen et al. [19] is carried out by a
were tested for stress conditions close to failure (σ1/σ1;f close or equal to
cyclic deviator stress of 24 kPa and a constant confining stress of 60 kPa.
one), which is outside the representative stress conditions of this study.
During the cyclic tests the stress ratio (σd/σc) is constant at 0.4, which is a
In this work, the parameters of Huurman’s model are adjusted so that
representative ratio in the subgrade of a full scale model test. It is noted that
the parameters εp0, a and B can be obtained; this adjustment took place
other confining stresses from 60 kPa to 210 kPa were also tested. The stress
through the formulation lscurvefit in Matlab.
path and the Mohr-Coulomb yielding criterion of each material are depicted in
In Rahman and Erlingsson [66], two silty sands are analyzed and
Fig. 7 showing the stress path applied on the cyclic triaxial test (with a stress
compared with the parameters of Chen’s model. The study also shows a
ratio equal to 0.4); and the failure envelopes of different materials used in
correlation between the εp0 and α parameters and the moisture content.
the laboratory tests. The failure envelopes are defined through the Mohr-
In the work developed by Chen et al. [19], a coarse sand and a silt are
Coulomb yielding criterion from the expression q = s + mp.
analysed and then the properties of these materials are divided ac-
It is seen that the distance between the stress path and the failure
cording to their granulometry (Table 4), their state conditions
envelopes influences the development of permanent deformation. In
(Table 5), and their failure parameters (Table 6).
fact, three models were selected, calibrated and compared (considering
It is important to first consider Huurman [38] work in order to
the long term response of the materials) to study and identify the main
understand the influence that granulometry (well and poorly graded
factors that influence the cumulative settlement: Huurman [38], Chen
soil) has on permanent deformation. The analysis utilises the following
et al. [19] and Rahman and Erlingsson [66]; these models have been
stress ratios: 0.695, 0.803, 0.911. Fig. 8 shows that well-graded sand
formulated and developed under different conditions in terms of ma-
has larger permanent deformation values for each stress ratio, which
terials and applied stress ratios, as described below.
means that granulometry has an important influence on permanent
With the Huurman [38] model, the samples are tested according to
deformation, as expected.
their optimum moisture content and with a confining pressure equal to

8
A. Ramos, et al. Transportation Geotechnics 23 (2020) 100326

Table 7 presents the calibrations of the models for each selected soil

optimum compaction – standard Proctor

optimum compaction - standard Proctor
optimum compaction - standard Proctor
optimum compaction - standard Proctor
where a regression analysis was performed. It is seen that there is some
correlation (exponential and polynomial) between the three constant
variables - εp0, α and β (superior to 0.7) - but correlation between the
field conditions and saturated
field conditions and saturated
constant variables and UIC classification (QS1, QS2, and QS3) is only
residual.
The comparison between the permanent deformation curves pre-
sented in Fig. 9. It is seen that the results of silty sand and coarse sand
(43% and 27% of fines) are very close, whereas the well and poorly-
graded sand and silt are at opposite sides (the highest and smallest
value of permanent deformation). This result is related to QS3-SW sand
because a lower value was expected in comparison to QS2-SP. Another
at
at
at
at
at
at
Observations

important fact is related to the influence of the stress ratio; in this in-
Compacted
Compacted
Compacted
Compacted
Compacted
Compacted

stance, the permanent deformation results are not 100% aligned with
those presented in Fig. 7 because higher permanent deformation is
expected when the stress path is close to the yielding criterion that de-
pends on the values s and m of the Mohr-Coulomb yielding criterion. This
means that in addition to the strength properties, the soil physical
properties (represented by the constants in the model) are also im-
portant in permanent deformation.
UIC classification

Ranking the materials

The materials and calibration process described in the previous section

QS2
QS1
QS1
QS1
QS2
QS3

enable the results to be interpreted and the materials can be ranked ac-
cording to their resistance to permanent deformation [22]. This informa-
tion can then be used to predict permanent deformation during the design
process. The ranking process provides reference values according to the
type of material as well as the moisture content close to optimum condi-
tions. In reality, Fig. 10 is a re-interpretation of Fig. 9. This innovative
ASTM Classification

procedure enables the expected values of permanent deformation to be

compared according to the type of material through the UIC and ASTM
soil classification; this ranking also includes different types of soils (clays
and sands) and different granulometry. Fig. 10 shows that the materials
CL-ML

classified as CL-ML have a higher permanent deformation than the sands

SW
SM
SM
SP

SP

(SM, SP, and SW), while the sands with fines have a higher deformation
than the SP and SW sands; therefore this parameter influences the per-
manent deformation and ranking of the geomaterials. During the design
process the long-term behaviour of the materials can be estimated, even
though the permanent deformation will be influenced by their state con-
Silty Sand (42.2% fines)
Silty Sand (27.4% fines)

ditions such as the moisture content; therefore the influence of this

parameter in this analysis must be traduced.
Poor-graded Sand
Well-graded Sand

Among other factors, moisture content also influences permanent

Materials selected for the preliminary analysis of the permanent deformation.

deformation. Puppala et al. [60] developed a model that depends on the

octahedral normal and shear stresses (Table A1 and Table A3). Different
Sand
Soils

materials, clay, silt, and sand with moisture contents ranging from dry
Silt

of optimum, optimum and wet of optimum were tested. Three different

confining pressures were applied to the clay and silt (0, 21 and 42 kPa),
and confining pressures of 21, 48 and 97 kPa were applied to the sand.
The deviatoric stress was determined by considering a percentage of the
maximum deviatoric stress of each soil sample obtained in the un-
Salour and Erlingsson [67] and Erlingsson et al. [29]

consolidated undrained triaxial tests, so the deviatoric stress varies

from 0.2σd,f, 0.4σd,f and 0.6σd,f.
To evaluate the influence that the moisture content has on perma-
nent deformation, the results from Puppala et al. [60] are used, but this
analysis only focuses on the behaviour of clay (QS1 – CL) because it was
not contemplated in previous studies. This exercise shows how im-
portant this parameter is, and the consequences in terms of permanent
deformation when the moisture content moves away from the optimum
conditions. Material properties are presented in Table 8, and as with the
previous models and materials, the curve of permanent deformation has
Chen et al. [19]

been adjusted to obtain the model parameters.

Huurman [38]

Fig. 11 shows that the differences between dry, optimum and wet
materials in terms of permanent deformation after 30 000 cycles is sig-
Author
Table 3

nificant; the optimum material is 4–8 times higher than the dry material,
and the wet material is 1–2 times higher than the optimum conditions.

9
A. Ramos, et al. Transportation Geotechnics 23 (2020) 100326

Fig. 7. Total stress path selected and failure envelopes of Mohr-Coulomb yielding criteria.

Table 4 estimation of the dynamic stresses and strategies to simulate the effect
Physical properties of the materials - granulometry. of rotation of principal stresses. Indeed, these topics will be evaluated in
Authors Huurman [38] Salour and Erlingsson Chen et al. [19]
further works related to the numerical modelling of a railway structure,
[67] and Erlingsson including the influence of the train’s speed in the stress results and
et al. [29] permanent deformation. It is expected that the train’s speed will in-
crease the stress amplitude (the stress path is close to the yielding cri-
Material/ C. Bruynweg Crusher Silty Clay Silty Clay Silt Coarse sand
terion) and also the permanent strain.
Properties (42.2% (27.4%
fines fines In this paper, it was found that permanent deformation could be
content) content) evaluated and predicted through laboratory tests, and most mechan-
istic-empirical models were based on cyclic triaxial tests.
Cu 2.1 10.5 28 33 2.51 4.68
Main concepts of permanent deformation and the laboratory tests used
Cc 1.05 1.25 0.54 0.75 1.32 0.62
D10 0.148 0.217 – – – – to predict them were reviewed, the approach used to simulate permanent
D30 0.219 0.784 – – – – deformation such as elastoplastic models, shakedown theory, and me-
D50 0.280 1.722 – – – – chanistic-empirical models were discussed. The definition of the shake-
D60 0.31 2.28 – – – – down is a very popular approach nowadays since it allows to prevent
CBR (%) 22 15.7 – – – –
plastic deformation that can be critical after some years of exploration. This
method became popular in pavement engineering with inclusion of 3D
Therefore, when the moisture content is far away from optimum condi- modelling and generalization of the method for anisotropic materials.
tions, the variations of permanent deformation in the clays are sub- However, the application of the shakedown analysis in railway engineering
stantial. In dry conditions, this value decreases significantly and in wet is limited. The shakedown approach however, demands the accurate as-
conditions, it increases which means that maximum permanent de- sessment of the response of the track foundation due to railway traffic,
formation occurs when the moisture content is beyond optimum (wet) which also demands powerful numerical modelling. It was found that the
due to an increase in the residual accumulation of excess water pressure. mechanistic-empirical model, such as the Chen’s model, reduced the time
needed for calculations, it has simple and elegant formulations, good pre-
dictive results, and is easy to implement in a numerical scheme. These
Conclusions mechanistic-empirical models were defined and categorised according to
the materials tested, the approach used in the laboratory (single-stage or
The ability to determine the permanent deformation of geomaterials multi-stage), the complexity of the formulation, and the variables included
is very important when modelling and evaluating the performance of in the mathematical expression. It was concluded that the permanent de-
pavements and railway structures; it is also a key factor when esti- formation models should also include the influence of the main stress
mating future maintenance operations and the respective costs. This variables (p and q), the initial state of the materials, the number of load
review paper has aimed to frame the main reasons why permanent cycles, and the distance to the rupture line; moreover, the constant para-
deformation models were developed, particularly in pavements and meters should also reflect the influence of factors such as the state condi-
railway structures. However, some issues related to traffic-induced tions of the material. However, those complex mechanistic-empirical
permanent deformation weren’t evaluated in this paper, such as the models that depend on several variables can also imply a need for tests

Table 5
State conditions of the materials.
Authors Huurman [38] Salour and Erlingsson [67] and Erlingsson et al. [29] Chen et al. [19]

Material/Properties C. Bruynweg Crusher Silty Clay (42.2% fines content) Silty Clay (27.4% fines content) Silt Coarse sand

3
γdry;max (kg/m ) 1723 1755 1998 2070 1620 2110
Wopt (%) 12.5 10.5 10.1 7.6 – –
γwet (kg/m3) 1942 1937 – – – –
Liquid limit (%) Non-plastic Non-plastic Non-plastic Non-plastic 35 Non-plastic
Plastic limit Non-plastic Non-plastic Non-plastic Non-plastic 24 Non-plastic

10
A. Ramos, et al. Transportation Geotechnics 23 (2020) 100326

Table 6
Strength properties of the materials.
Authors Huurman [38] Salour and Erlingsson [67] and Erlingsson et al. [29] Chen et al. [19]

Material/Properties C. Bruynweg Crusher Silty Clay (42.2% fines content) Silty Clay (27.4% fines content) Silt Coarse sand

o
ϕ( ) 48.2 50.20 36.18 45.66 11.7* 0*
c (kPa) 5.60 8.68 15.82 10.43 16.4* 33*
s (kPa)** 9.90 14.90 31.80 19.10 24.8 0
m** 1.98 2.07 1.47 1.88 0.62 1.33

* In this case, the author described the saturated samples (silt and coarse sand) in ϕ’ and c’
** s and m are the parameters used to define the failure envelope: q = s + mp

other than a standard triaxial cyclic test, in order to obtain more para-
meters.
The mechanistic-empirical models summarised in the Annex were
divided according to the type of material (clay, silt, sand and gravel) to
better understand the model’s formulation and the condition of the
material tested in terms of its granulometry (Cu and Cc) and plasticity
properties (WL, WP, and IP). The Tables also include some observations
regarding the laboratory tests and classification of the material (UIC
and ASTM). This information (including the classification) can be very
helpful when modelling sub-structure and predicting its performance.
Considering the available mechanistic-empirical models as well as
understanding the main variables and factors that can affect the long-
term response of geomaterials, it is important to evaluate the robustness
of the models and their sensitivity. Thus, some models and materials
Fig. 8. Comparison between the permanent deformation of a poorly graded (C. were selected to perform a parametric study. The choice depends on the
Bruynweg) and well-graded (crusher) sand under different stress ratios. data available for each material and the diversity and variability of the
geomaterials (type of soils) in terms of its granulometry, percentage of
Table 7 fines, moisture content, and plasticity properties, etc.
Parameters of Chen’s model and its relationship with the ASTM and UIC clas- In the comparison analysis: silt, silty sand, and sand were divided
sification. according to properties such as the percentage of fines (27% and 43%)
εp0 α β ASTM UIC classification and granulometry (poor and well-graded materials). Selected experi-
classification mental results found in the literature were then adapted to Chen’s
model. The calibration process was performed to compare the long-
Silt [19] 0.0029 0.2680 0.3390 CL QS1
term response of these geomaterials, after which they were defined and
Silty sand (43% 0.0067 0.6500 0.2000 SM QS1
fines) [28] ranked based on the results of permanent deformation. This ranking
Silty sand (27% 0.0069 0.6500 0.2000 SM QS1 tool can be used during the design process for estimating the permanent
fines) [28] deformation of a certain type of material after N loading cycles, while
Coarse sand [19] 0.0018 0.3340 0.3390 SP QS2 considering the ASTM and UIC classification. As expected, the com-
Poor-graded sand 0.0505 0.0104 0.0018 SP QS2
parison showed that the values of permanent deformation decreased
[38]
Well-graded sand 0.0486 0.0111 0.0569 SW QS3 according to the UIC classification, so well-classified materials showed
[38] a reduction in permanent deformation. Furthermore, the deformation
behaviour of these materials also depended on their granulometry, the

Fig. 9. Permanent deformation of the materials applying the Chen et al. [19] model.

11
A. Ramos, et al. Transportation Geotechnics 23 (2020) 100326

Fig. 10. Permanent deformation predicted by Chen et al. [19]’s model for different soils (N = 30,000 cycles) under a stress ratio σ1/σ1:f = 0.4.

Table 8
Properties of the clay (in optimum conditions) tested in the Puppala
et al. [60]
Properties and parameters Clay

o
ϕ( ) 18
c (kPa) 60
Liquid limit (%) 28.19
Plasticity index (%) 12.55
Maximum dry unit weight (kN/m3) 17.10
Wopt (%) 17.11
α1 0.24
α2 0.31
α3 −1.77
α4 2.85

percentage of fines, and the moisture content. Based on the proposed

ranking, a brief analysis has been carried out to understand and
quantify the influence of the moisture content. The results showed that
for a material classified as QS1 (CL), designated as clay, the moisture
content exerts a significant influence on permanent deformation.

CRediT authorship contribution statement

Ana Ramos: Writing - original draft. António Gomes Correia:

Conceptualization, Methodology, Supervision, Funding acquisition,
Writing - review & editing. Buddhima Indraratna: Writing - review &
editing. Trung Ngo: Writing - review & editing. Rui Calçada: Writing -
review & editing. Pedro Alves Costa: Writing - review & editing.

Acknowledgments

This work was partially carried out under the framework of

In2Track, a research project of Shift2Rail. It was also supported by FCT Fig. 11. Influence of the moisture content in the permanent deformation of a
clay (Puppala’s model): (a) permanent deformation results applying the Chen
- “Fundação para a Ciência e a Tecnologia” - PD/BD/127814/2016. The
et al. [19] model; (b) Ranking of permanent deformation for clay.
Authors also wish to acknowledge the collaboration among them
through the scheme of Industrial Transformation Training Centre for
advancing track infrastructure (ITTC-Rail, IC170100006), funded by
the Australian Research Council.

Annex

An extensive list of the permanent deformation based on laboratory tests is presented in this Annex.
Although these tables are constructed in a uniform manner, especially the nomenclature of the variables and the parameters and characteristics of
the materials, the authors still maintained the original symbols and nomenclature describe in the original paper.

12
 Table A1
Summaries of permanent deformation models for fine-grained soils - clays.
UIC ASTM classification Author’s model Equation Variables and Empirical Constants Parameters and characteristics Observations
A. Ramos, et al.

Model

– – Monismith et al. [54] p = AN b - εp is the plastic deformation (%); – - This model has a limited practical
- N is the number of load cycles; used since there are other factors,
- A and b are two parameters that besides N, that can influence the
represent the influence of other permanent deformation;
factors. - b is an independent parameter on the
deviatoric stress;
- The coefficient A corresponds to the
plastic deformation after the first cyclic
load.
– – Chai and Miura [11] m n b - qs is the initial static deviatoric – - This model includes the effects of
p qd qs
=a 1+ N stress; initial static stress, the magnitude and
qf qf
- n = constant. number of applied loads associated
with the passage of the vehicle, and
the properties of the subsoil in terms
of compression and strength.
– – Yesufa and Hoffa = N m (1 ( d qs ) - p1, N is the axial permanent – - The parameters of these materials
p1, N p1,0
[85]* deformation for the cyclic load N; were found through triaxial tests: the
- p1,0 is the cumulative axial static deviatoric stress comes from the
permanent deformation after the first difference between the axial and
cyclic load; confining stress in failure;
- σd is the deviatoric stress induced by - p1,0 is from the equivalent static load in
the passage of the vehicle; the first cyclic load (based on the
- qs is the static stress measured on the plasticity theory).
soil;

13
- m is a constant determined to fit the
laboratory test’s results.
oct 3 oct 4
– – Cai et al. [10] p - α1, α2, α3, α4 and, α5 are constants of – - This model is based on a series of tests
z = 1N 2 ( 1+4 2) 5
atm
( ) ( ) atm the model and determined through using the hollow cylinder apparatus.
laboratory tests; The characteristics of the permanent
- σoct and τoct are the normal and shear deformation of a sand were then
octahedral stresses, respectively; studied in drained conditions by
- σatm is the reference stress; considering different confining stress
- η is the cyclic torsional stress ratio. levels, ratio of the cyclic vertical stress
and ratio of the cyclic torsional stress;
- The model was developed to include the
rotation of principal stresses.
QS0High plasticity CHFat clay Li and Selig [49] m - qd is the traffic-load-induced - a = 0.64;
p qd
=a Nb dynamic deviator stress; - b = 0.10;
qf
- qf is the static failure deviator stress - m = 1.7.
of soil;
- a, b and m are soil parameters and
are related to the plasticity index of
the subsoil.
QS0Heavy clay CHFat Clay Puppala et al. [59] p oct - A, β and α are soil parameters; - WL = 93%; - The model was tested based on triaxial
log = logA + logN + log ( ) atm - σoct and τoct are the octahedral normal - WP = 29%; tests;
and shear stresses, respectively. - IP = 64% (high plasticity); - During this study, materials such as well
- Sand content = 0%; graded sand, silty clay and clay were
- Silt content = 5%; considered.
- Clay content = 95%.
(continued on next page)
Transportation Geotechnics 23 (2020) 100326
 Table A1 (continued)

UIC ASTM classification Author’s model Equation Variables and Empirical Constants Parameters and characteristics Observations
Model
A. Ramos, et al.

QS0WL = 60%High (without enough Wei et al. [76] – q, r = AN b ; A = q, r = K q KT K K - q, r is the axial residual strain; - Wn = 60–62%; - This model is based on the power
plasticity data to carry out the based onGuo [36] - A is the residual strain generated in - Gs = 2.71; St = 5.9; function developed by Monismith
ASTM classification) -(Wenzhou structural the first cycle; - M (critical stress et al. [54],
clay) - b is a material parameter ratio) = 1.31. - The parameter A depends on the initial
determining the cumulating rate of stress state and the imposed dynamic
the residual strain with loading stress on non-remoulded Wenzhou clay
cycles and is equal to the gradient of samples [36]. The samples were
the residual strain curve in the ln q, r collected from a depth of 6–7 m;
– ln N coordinate; - The parameters are based on the work
- K q and KT are functions to reflect the by Xiao et al. [81] who carried out a
effect of wheel load-induced stress; series of tests based on the hollow
- K and K are functions to reflect the cylinder apparatus.
effect of initial static stress state and
dynamic stress combinations.
QS1(low plasticity) CLLean clay Hyde [39] p q - a is a constant; - WL = 32%; - The tests included a constant and
1 =a
3 - σ3 is the confining stress; - WP = 18%; variable confining stress;
- q is deviator stress. - IP = 14%: - During the analysis, the behaviour of the
- Gs = 2.74. marl was studied;
- The author states that these results
would be the same if medium confining
stress was applied.
QS1Low to medium CLLean clay Li and Selig [49] m - qd is the traffic-load-induced - a = 0.84;
p qd
plasticity =a Nb dynamic deviator stress; - b = 0.13;
qf
- qf is the static failure deviator stress - m = 2.0.
of soil;

14
- a, b and m are soil parameters and
are related to the plasticity index of
the subsoil.
QS1Silty clay CLLean clay Puppala et al. [59] p oct - A, β and α are soil parameters; - WL = 44%; - The model was tested based on triaxial
log = logA + logN + log
atm
( ) - σoct and τoct are the octahedral - WP = 15%; tests;
normal and shear stresses, - IP = 29% (medium - During this study, well-graded sand,
respectively. plasticity); silty clay and clay were tested.
- Sand content = 10%;
- Silt content = 70%;
- Clay content = 20%.
QS1Silty clay CLLean clay Puppala et al. [60] 3 4 - α1, α2, α3 and α4 are the model - WL = 28.19%; - The model is based on 10,000 cycles;
p oct oct
= 1N 2 constants determined from - IP = 12.55% (low - It includes the effects of mean and shear
Pref Pref
laboratory tests; plasticity); stresses;
- σoct and τoct are the octahedral normal - Gs = 2.63; - The model was tested based on the
and shear stresses, respectively; - Passing #200 = 80%; cyclic triaxial tests;
- Pref is the reference stress. - Maximum dry unit - The permanent deformation of clay, silt
weight = 17.10 kN/m3; and sand were measured from samples
- Wopt = 17.11%; of compacted soil with different water
- c = 60 kPa; content.
- ϕ = 18°.
QS1Silty clayed sand CLLean Clay Puppala et al. [60] 3 4 - α1, α2, α3 and α4 are the model - WL = 16.70%; - The model is based on 10,000 cycles;
p oct oct
= 1N 2 constants determined from - IP = 7.50% (low plasticity); - It includes the effects of the mean and
Pref Pref
laboratory tests; - Gs = 2.70; shear stresses;
- σoct and τoct are the octahedral normal - Passing #200 = 38%; - The model was tested based on the
and shear stresses, respectively; - Maximum dry unit cyclic triaxial tests;
- Pref is the reference stress. weight = 16.9 kN/m3; - The permanent deformations of clay, silt
- Wopt = 19.3%; and sand were measured from samples
- c = 103 kPa,; of c soil compacted with different water
- ϕ = 35°. contents.
Transportation Geotechnics 23 (2020) 100326

(continued on next page)

 Table A1 (continued)

UIC ASTM classification Author’s model Equation Variables and Empirical Constants Parameters and characteristics Observations
Model
A. Ramos, et al.

QS1 CLLean clay Wei et al. [76] – q, r = AN b ; A = q, r = K q KT K K - q, r is the axial residual strain; - IP = 21.8%; - The model is based on the power
based onXiao et al. - A is the residual strain generated in - IL = 1.35 (medium function developed by Monismith
[81] -Clay (Shanghai the first cycle; plasticity); et al. [54],
structural soft clay) - b is a material parameter - Gs = 2.74; - The parameter A depends on the initial
determining the cumulating rate of - Wn = 51.8%; stress state and the imposed dynamic
the residual strain with loading - e0 = 1.402; stress on samples of non-remoulded
cycles and is equal to the gradient of - M (critical stress Wenzhou clay [36]. The samples were
the residual strain curve in the ln q, r ratio) = 1.28. collected from a depth of 6–7 m;
– ln N coordinate; - The parameters were determined based
- K q and KT are functions to reflect the on the work by Xiao et al. [81] who
effect of wheel load-induced stress; performed a series of tests based on the
- K and K are functions to reflect the hollow cylinder apparatus.
effect of initial static stress state and
dynamic stress combinations.

15
Table A2
Summarises of permanent deformation models for fine-grained soils - silts.
UIC ASTM Author’s Equation model Variables and Empirical Constants Parameters and characteristics Observations
classification model

QS1Low to MLSilt Li and Selig m - qd is the traffic-load-induced - a = 1.2;

p qd
medium [49] =a Nb dynamic deviator stress; - b = 0.18;
qf
plasticity - qf is the static failure deviator - m = 2.4.
stress of soil;
- a, b and m are soil parameters and
are related to the plasticity index
of the subsoil.
QS1 ML-CLSilt - Lean Chen et al. a - 1p (N ) is the strain settlement; - WL = 35%; - The model is based on
p p0 pam2 + qam2 1
clay [19] 1 (N ) = 1 [1 e BN ]
- 1p0 , B, a, s and m are material - WP = 24%; Gidel’s model [30],
pa p s (qini + qam )
m 1 + ini + parameters; - IP = 11% (low to medium - This model takes into
pam pam pam
- pa = 100 kPa; plasticity); account the direct
- pini and qini are the mean and - Gs = 2.67; influence of the initial
deviator stress in the initial state of - Cu = 2.51;Cc = 1.32; stress state.
the layer; - Maximum dry density = 15.89 kN/
- pam and qam are the amplitude of m3;
mean stress and deviator stress for - Hydraulic
train loadings, respectively; permeability = 5.3 × 10-7 m/s;
- s is the intersection of the Mohr- - c’=11.7 kPa;
Coulomb yielding criterion in the q - ϕ’=16.4°.
axle in the p-q space;
- m is the slope of the yielding
criterion.
Transportation Geotechnics 23 (2020) 100326
 Table A3
Permanent deformation models for granular soils (sands).
UIC ASTM Author’s Equation model Variables and Empirical Constants Parameters and characteristics Observations
A. Ramos, et al.

classification model

– – MEPDG 0 ( N) - βs is the global calibration – - This model

p = s v
model [4] r
( )e coefficient; considers the
- εr is the resilient strain effect of the stress
imposed in the laboratory test; on permanent
- εv is the average vertical deformation by
resilient strain (in the base linearly
layer of the flexible projecting the
pavements, for example). plastic
deformation
obtained from the
laboratory tests
through vertical
strains. This
projection is an
assumption
without any
experimental or
mechanical
justification; it is
therefore
inaccurate due to
the nonlinear
effect that stress
has on the

16
permanent
deformation of
unbound granular
materials – UGM
[35].
– – Korkiala- R - C is a parameter dependent on – - This model is
p (N ) = CN b
Tanttu [44] A R the stress and also compaction based on the
degree and water content; Sweere’s model
- b is a parameter dependent on [69].
the stress level, ratio of failure, - It uses the
compaction degree and water deviatoric stress
content; ratio to capture the
- A corresponds to the nonlinear effect of
maximum value of the ratio the stress state.
(the author suggests a value However, plastic
close to 1.05); deformation is
q
- R = , q is the deviatoric infinity when the
qf
stress and qf corresponds to load cycles go to
the failure deviatoric stress; infinity, so the
model cannot
demonstrate the
hardening and
softening behavior
of UGM.
– – Chow et al. f D - σd is the deviatoric shear – - The model
p = AN B C
[20] d ( ) max stress; includes the
- τf is the shear stress; power functions
- τmax is the shear strength; of deviatoric
- A, B, C and D are the shear stress and
regression coefficients. the shear strength
Transportation Geotechnics 23 (2020) 100326

ratio;
(continued on next page)
 Table A3 (continued)

UIC ASTM Author’s Equation model Variables and Empirical Constants Parameters and characteristics Observations
classification model
A. Ramos, et al.

- The author
developed 16 types
of materials with
one confining
pressure
(=34.5 kPa) and
three deviatoric
stresses. The RLT
tests show that this
model has higher R2
values but when the
number of load
cycles (N) goes to
infinity, the
accumulated plastic
deformation also
goes to infinity at
one confining
pressure. Moreover,
the deviatoric shear
stress term affects
the shear ratio in
the model and both
parameters
represent the

17
softening behavior
of the material
without including
the hardening
phenomenon.
– – Xiao et al. b c - θ is the sum of the principal – - Triaxial cyclic
f f
p =a N d EXP e (S Sopt ) + f wPI + g wPI + h (S Sopt )
[82] p0 max max p0 stresses; tests took place
- τf/τmáx is the ratio related to on the limestone,
the shear stress; dolomite and
- S corresponds to the saturation gravel samples
level; (materials used in
- Sopt corresponds to the base, sub-base
optimum saturation level; and subgrade
- wPI corresponds to a weighted layers treated in
plasticity index; Illinois);
- a, b, c, d, e, f, g and h are the - The cohesion and
parameters determined based friction angle were
on regression analysis. determined for
three confining
stresses.
QS1Silty sand with SMSilty sand Rahman q - q is deviatoric stress; - Cu 28; Cc 0.54; - The triaxial tests
high fine and pa - p is mean stress; - Gs = 2.68; were carried out
p (N ) = N bSf Sf Sf =
content Erlingsson p - pa atmospheric pressure; - Fine content = 42.2%; by considering
[66] pa - a, b and α are the regression - Maximum dry density = 19.6 the constant
andSalour parameters. kN/m3; confining stress;
and - Wopt = 10.1%. - They considered
Erlingsson high stress levels
[67]
(continued on next page)
Transportation Geotechnics 23 (2020) 100326
 Table A3 (continued)

UIC ASTM Author’s Equation model Variables and Empirical Constants Parameters and characteristics Observations
classification model
A. Ramos, et al.

QS1Silty sand with SMSilty sand Rahman q - q is deviatoric stress; - Cu 33; Cc 0.75; and reduced stress
high fine and pa - p is mean stress; - Gs = 2.67; levels.
p (N ) = N bSf Sf Sf =
content Erlingsson p - pa atmospheric pressure; - Fine content = 27.4%;
[66] pa - a, b and α are the regression - Maximum dry density = 20.3
andSalour parameters. kN/m3;
and - Wopt = 7.6%.
Erlingsson
[67]
QS2 SPPoorly Huurman p N B N - A, B, C, D are parameters - Cu = 1.69; Cc = 1.39; - Cyclic triaxial
1 (N ) =A + C eD 1000 1
graded sand [38] 1000
( ) function of stress level and - Maximum dry Density = 16.68 tests were carried
stress ratio (σ1/σ1;f). kN/m3; out on the sands
- Wopt = 12.5%; used in the base
- c = 4.08 kPa; ϕ = 43.90. layers;
QS2 SPPoorly Huurman p N B N - A, B, C, D are parameters - Cu = 1.69; Cc = 0.998; - The samples were
1 (N ) =A + C eD 1000 1
graded sand [38] 1000
( ) function of stress level and - Maximum dry Density = 16.73 subjected to a
stress ratio (σ1/σ1;f). kN/m3; confining stress
- Wopt = 13.0%; equal to 12 kPa
- c = 6.76 kPa; ϕ = 43.00. (corresponds to a
QS2 SPPoorly Huurman p N B N - A, B, C, D are parameters - Cu = 1.88; Cc = 1.04; depth equal to
1 (N ) =A + C eD 1000 1
graded sand [38] 1000
( ) function of stress level and - Maximum dry Density = 16.56 0.60 m);
stress ratio (σ1/σ1;f). kN/m3 - During the test, a
- Wopt = 14.0%; million cycles were
- c = 6.34 kPa; ϕ = 41.80; applied (frequency
QS2 SPPoorly Huurman p N B N - A, B, C, D are parameters - Cu = 2.10; Cc = 1.05; equal to 5 Hz).
1 (N ) =A + C eD 1000 1
graded sand [38] 1000
( ) function of stress level and - Maximum dry Density = 16.90

18
stress ratio (σ1/σ1;f). kN/m3;
- Wopt = 12.5%;
- c = 5.60 kPa; ϕ = 48.20;
QS2 SPPoorly Huurman p N B N - A, B, C, D are parameters - Cu = 1.70; Cc = 1.10;
1 (N ) =A + C eD 1000 1
graded sand [38] 1000
( ) function of stress level and - Maximum dry Density = 15.61
stress ratio (σ1/σ1;f) kN/m3;
- Wopt = 14.50%.
- c = 7.19 kPa; ϕ = 42.80;
QS2 SPPoorly Huurman p N B N - A, B, C, D are parameters - Cu = 2.35; Cc = 1.14; - Cyclic triaxial
1 (N ) =A + C eD 1000 1
graded sand [38] 1000
( ) function of stress level and - Maximum dry Density = 16.16 tests took place
stress ratio (σ1/σ1;f). kN/m3; on the sands used
- Wopt = 14.5%; in the base layers;
- c = 7.99 kPa; - The samples were
- ϕ = 39.70. subjected to a
QS2 SPPoorly Huurman p N B N - A, B, C, D are parameters - Cu = 3.76; Cc = 1.37; confining stress
1 (N ) =A + C eD 1000 1
graded sand [38] 1000
( ) function of stress level and - Maximum dry Density = 16.16 equal to 12 kPa
stress ratio (σ1/σ1;f). kN/m3; (corresponds to a
- Wopt = 13.5%; depth equal to
- c = 7.48 kPa; 0.60 m);
- ϕ = 42.90; - During the test, a
million cycles were
applied (frequency
equal to 5 Hz).
(continued on next page)
Transportation Geotechnics 23 (2020) 100326
 Table A3 (continued)

UIC ASTM Author’s Equation model Variables and Empirical Constants Parameters and characteristics Observations
classification model
A. Ramos, et al.

QS2 SPPoorly Sweere [69] log ( 1p (N )) = a + blog (N ) - a,b – material parameters - Density = 16.79 kN/m3; - Application of 106
graded sand - for each stress level, separate - Cu = 1.68; cyclic loads
permanent strain parameters a - Cc = 1.10. during the triaxial
and b need to be determined cyclic tests;
from the results of the cyclic - The samples
load triaxial test. consisted of
- **The Sweeres’s formula is granular materials
valid for granular materials used in the base
[30]. layers, and sands;
- The UGM samples
were subjected to
triaxial cyclic tests.
QS2Sa (> 0.063 to SPPoorly Puppala 3 4 - α1,α2,α3 and α4 are the model - WL = 26.40%; - The model is based
p oct oct
2) graded sand et al. [60] = 1N 2 constants determined from - Gs = 2.71%; on 10,000 cycles,
Pref Pref
laboratory tests; - Passing #200 = 0.70%; and includes the
- σoct and τoct are the octahedral - Cu = 1.79; Cc = 0.89; effects of mean and
normal and shear stresses, - Maximum dry unit shear stresses;
respectively; weight = 15.70 kN/m3; - The model was
- Pref is the reference stress. - Wopt = 13.70%; tested based on the
- c = 20 kPa; cyclic triaxial tests;
- ϕ = 42°; - The permanent
deformation of clay,
silt and sand were
measured using the
soil samples

19
compacted with
different water
content.
a p
QS2Gravelly sand SPSand Chen et al. - 1 (N ) is the strain settlement; - Gs = 2.66; - The model is
p p0 pam2 + qam2 1 p0
(SaGr) poorly [19] 1 (N ) = 1 [1 e BN ]
- - Cu = 4.8; Cc = 0.62; based on Gidel’s
pa p s (qini + qam ) 1 , B, a, s and m are material
graded m 1 + ini + parameters; - Maximum dry density = 20.69 model [30],
pam pam pam
- pa = 100 kPa; kN/m3; - It considered the
- pini and qini are the mean and - Minimum dry density = 15.89 direct influence of
deviator stress in the initial kN/m3; the initial stress
state of the layer; - Hydraulic state.
- pam and qam are the amplitude permeability = 3.2 × 10-5m/s;
of mean stress and deviator - c’=0 kPa;
stress for train loadings, - ϕ’=33°.
respectively.
oct 3 oct 4
QS2Sa (> 0.063 to SPPoorly Cai et al. p - α1,α2,α3,α4 and α5 are - Gs = 2.70; - The model is based
z = 1N 2 ( 1+4 2) 5
2)Fine sand graded sand [10] atm
( ) ( ) atm constants of the model and - Cu 1.63; on a series of tests
determined through - Cc 0.89; using the hollow
laboratory tests; - D50 = 0.17 mm; cylinder apparatus.
- σoct and τoct are the normal and - emax = 1.142;emin = 0.628. The permanent
shear octahedral stresses, deformation of a
respectively; sand was studied
- σatm is the reference stress in drained
- η is the cyclic torsional stress conditions by
ratio. considering
different levels of
confining stress,
and the ratios of
cyclic vertical
stress and cyclic
torsional stress;
Transportation Geotechnics 23 (2020) 100326

(continued on next page)

 Table A3 (continued)

UIC ASTM Author’s Equation model Variables and Empirical Constants Parameters and characteristics Observations
classification model
A. Ramos, et al.

- This model was

developed to
incorporate the
rotation of principal
stresses.
QS2 SPPoorly- Lekarp and p b - a and b: - Dry density = 14.91 kN/cm3; - Triaxial cyclic
1 (Nref ) q
=a
graded sand Dawson (L p0 ) () p - Nref; - Cu 1.5; tests and hollow
max
[46] - L = q2 + p2 ; - Cc 1.0. cylinder tests
- p0 = 100kPa (reference mean were carried out
stress). on limestone,
sand and gravel.
N acc th
QS2 SPPoorly- Qian et al. acc - N is the N vertical - Gs = 2.54; - Drained cyclic
N =( 1 + 2) +( 3 + 4 ) ln 100
graded sand [63] accumulative strain in - Cu 1.65; triaxial tests with
percentage; - Cc 1; different levels of
- Nisthenumberofloadingcycles - D50 = 0.16 mm initial mean
- is the calculated shear - emax 0.931; effective stress
strain amplitude in percentage - emin 0.611. (ps), relative
- 1 4 are model parameters density (Dr),
initial static stress
ratio (ns) and
cyclic stress ratio
were performed
on saturated
Toyoura sand.
Indeed, the effects

20
of cyclic stress
ratio and initial
mean effective
stress on the
permanent
deformation;
- The estimated
permanent strains
show a good
agreement with the
measured data. The
deviation between
the estimation and
the measurement
is < 20%.
QS3 SPPoorly Hyde [39] p q - a is a constant; - Gs = 2.72; - The tests were
1 =a
graded sand 3 - σ3 is the confining stress; - Cu 16; carried out under
- q is deviator stress. - Cc 0.69; a constant and
- Maximum dry densisty = 21.76 variable confining
kN/m3; stress;
(continued on next page)
Transportation Geotechnics 23 (2020) 100326
 Table A3 (continued)

UIC ASTM Author’s Equation model Variables and Empirical Constants Parameters and characteristics Observations
classification model
A. Ramos, et al.

- Wopt = 7.5 ± 0.5%; - During the analysis,

the behaviour of the
marl was studied;
- The author states
that the results
would be the same
if medium confining
stress was applied.
QS3 SPPoorly- Lekarp and p b - a and b: - Dry density = 20.10 kN/m3; - Triaxial cyclic
1 (Nref ) q
=a
graded sand Dawson (L p0 ) p
() - Nref; - Cu 32; tests and hollow
max
[46] - L = q2 + p2 ; - Cc 0.2. cylinder tests
- p0 = 100kPa (reference mean were carried out
stress). on limestone,
sand and gravel.
QS3Sa(> 0.063 to SWWell- Puppala p oct - A, β and α are soil parameters; - % sand = 100. - The model was
log = logA + logN + log
2) graded sand et al. [59] ( ) atm - σoct and τoct are the octahedral tested based on
normal and shear stresses, triaxial tests;
respectively. - During his study,
well-graded sand,
silty clay and clay
were considered.
QS3 SWWell- Huurman p N B N - A, B, C, D are parameters - Cu = 10.5; Cc = 1.25; - Cyclic triaxial
1 (N ) =A + C eD 1000 1
graded sand [38] 1000
( ) function of stress level and - Maximum dry Density = 17.21 tests were carried
stress ratio (σ1/σ1;f) kN/m3; out on the sands
- Wopt = 10.5%; used in the base

21
- c = 8.68 kPa; layers;
- ϕ = 50.20. - The samples were
subjected to a
confining stress
equal to 12 kPa
(corresponds to a
depth equal to
0.60 m);
- During the test, a
million cycles were
applied (frequency
equal to 5 Hz).
Transportation Geotechnics 23 (2020) 100326
 Table A4
Permanent deformation model for granular soils (gravels and sandy gravels).
UIC ASTM Author’s Equation model Variables and Empirical Constants Parameters and characteristics Observations
A. Ramos, et al.

classification model

QS2Sandy gravel GP-GM/GCPoorly- Rahman and q - q is deviatoric stress; - Gs = 2.54; - The triaxial tests took place by
(GrSa)Cu 200; Cc 0.8 graded sandy Erlingsson pa - p is mean stress; - Maximum dry considering constant confining
p (N ) = N bSf Sf Sf =
gravel(no [66] p - pa atmospheric pressure; density = 21.67 kN/m3; stress;
information about pa - a, b and α are the regression - Fines Content = 10.2%; - The tests considered high stress
the fines) parameters. - Wopt = 6%. levels and reduced stress levels.
QS2Sandy gravel GW-GM/GCWell- Rahman and q - q is deviatoric stress; - Gs = 2.68;
(GrSa)Cu 73; Cc 1.5 graded sandy Erlingsson pa - p is mean stress; - Fines Content = 8.6%;
p (N ) = N bSf Sf Sf =
gravel(no [66] p - pa atmospheric pressure; - Maximum dry density
information about pa - a, b and α are the regression - = 23.05 kN/m3;
the fines) parameters. - Wopt = 6.9%.
QS2Sandy gravel GW-GM/GCWell- Rahman and q - q is deviatoric stress; - Gs = 2.64;
(GrSa)Cu 98; Cc 2.8 graded sandy Erlingsson bS pa - p is mean stress; - Fines Content = 6.5%;
p (N ) = N f Sf Sf =
gravel(no [66] p - pa atmospheric pressure; - Maximum dry
information about pa - a, b and α are the regression density = 22.16 kN/m3;
the fines) parameters. - Wopt = 6%.
QS2Sandy gravel GW-GM/GCWell- Rahman and q - q is deviatoric stress; - Gs = 2.64;
(GrSa)Cu 117; Cc 2.2 graded sandy Erlingsson pa - p is mean stress; - Fines Content = 12%;
p (N ) = N bSf Sf Sf =
gravel (no [66] p - pa atmospheric pressure; - Maximum dry
information about pa - a, b and α are the regression density = 21.77 kN/m3;
the fines) parameters. - Wopt = 6.5%.
QS2-QS3 (average to good – Gidel et al. B u 1 - ε0, B and u are material - 0/20 mm (limestone - This model is based on cyclic
0 N Lmax s qmax
p (N ) = 1 × m+
soils)Classification based [30] 100
( ) pa pmax pmax parameters unbound granular triaxial tests on two samples
on the LA and MDE - pmax and qmax are the maxima material); where the particles had
results. applied hydrostatic stress and - LA = 22; MDE = 15; different dimensions: hard

22
the deviator stress, - pdSPO = 2450 kg/m3; limestone (0/20 mm) and
respectively. - WSPO = 4.4%; microgranite (0/10 mm);
- s is the intercept of the Mohr- - Fines content = 10%; - The tests included several load
Coulomb failure line in the p-q - D = 160 mm; levels (40) at almost 20,000
space H = 320 mm. cycles.
- m is the slope of this failure line
- pa correspond to the reference
stress (100 kPa).
- The value Lmax is calculated
considering the following
expression:
- Lmax = pmax 2 + qmax 2
QS2-QS3(average to good – Gidel et al. B u 1 - ε0, B and u are material - 0/10 mm (granular
0 N Lmax s qmax
p (N ) = 1 × m+
soils)Classification of the [30] 100
( ) pa pmax pmax parameters material obtained from
subgrade (based on the LA - pmax and qmax are the maxima microgranite);
and MDE results): applied hydrostatic stress and - LA = 20; MDE = 13;
the deviator stress, - pdSPO = 2200 kg/m3;
respectively. - WSPO = 6.3%
- s is the intercept of the Mohr- - Fines content = 7.8%;
Coulomb failure line in the p-q - D = 76.2 mm.
space
- m is the slope of this failure line
- pa correspond to the reference
stress (100 kPa).
- The value Lmax is calculated
considering the following
expression:
- Lmax = pmax 2 + qmax 2
(continued on next page)
Transportation Geotechnics 23 (2020) 100326
 Table A4 (continued)

UIC ASTM Author’s Equation model Variables and Empirical Constants Parameters and characteristics Observations
classification model
A. Ramos, et al.

QS2 – QS3Sandy gravel (GrSa) GWWell-graded Ling et al. acc (N ) = fampl fn fp fN - fampl describes the range - Cu 60; - The study focused on granular
gravel [50] intensity of PD with the - Cc 1.3; materials.
increase of the cyclic stress - Sand content = 40%;
amplitude; - Gravel content = 60%;
- fp represents the influence of
the initial mean stress;
- fn represents the influence of
the initial stress ratio;
- fN characterizes the trend of the
variation’s deformation with
the number of cyclic loads.
QS3 GPPoorly-graded Lekarp and p b - a and b ; - Dry density = 21.18 - Triaxial cyclic tests and hollow
1 (Nref ) q
=a
gravel Dawson [46] (L p0 ) () p - Nref; kN/m3; cylinder tests took place on
max
- L = q2 + p2 ; - Cu 92; limestone, sand and gravel.
- p0 = 100kPa (reference mean - Cc 6.4.
stress).
QS3Sandy gravel GPPoorly-graded Rahman and q - q is deviatoric stress; - Gs = 2.49; - The triaxial tests considered
(GrSa)Cu 26; Cc 0.9 sandy gravel Erlingsson bS pa - p is mean stress; - Fines Content constant confining stress;
p (N ) = N f Sf Sf =
[66] p - pa atmospheric pressure; (< 0.075 mm) = 3.8%; - The tests also considered high
pa - a, b and α are the regression - Wopt = 7.5%; stress levels and reduced stress
parameters. - Maximum dry levels.
density = 20.89 kN/m3.
QS3Sandy gravel GPPoorly-graded Rahman and q - q is deviatoric stress; - Fines Content
(GrSa)Cu 35; Cc 4.6 sandy gravel Erlingsson bS pa - p is mean stress; (< 0.075 mm) = 1.4%;
p (N ) = N f Sf Sf =
[66] p - pa atmospheric pressure; - Wopt = 5%;

23
pa - a, b and α are the regression - Gs = 2.61;
parameters. - Maximum dry
density = 21.18 kN/m3.
QS3 GWWell-graded Lekarp and p b - a and b; - Dry density = 20.50 - Triaxial cyclic tests and hollow
1 (Nref ) q
=a
Gravel Dawson [46] (L p0 ) () p - Nref; kN/m3; cylinder tests took place on
max
- L = q2 + p2 ; - Cu 75; limestone, sand and gravel.
- p0 = 100kPa (reference mean - Cc 1.3.
stress);
QS3 GWWell-graded Lekarp and p b - a and b ; - Dry density = 22.16
1 (Nref ) q
=a
Gravel Dawson [46] (L p0 ) () p - Nref kN/m3;
max
- L = q2 + p2 ; - Cu 55;
- p0 = 100kPa (reference mean - Cc 1.3.
stress).
QS3Sandy gravel GWWell-graded Rahman and q - q is deviatoric stress; - Gs = 2.64; - The triaxial tests considered
(GrSa)Cu 16; Cc 1.6 sandy gravel Erlingsson pa - p is mean stress; - Fines Content = 2.4%; constant confining stress;
p (N ) = N bSf Sf Sf =
[66] p - pa atmospheric pressure; - Maximum dry - The tests also considered high
pa - a, b and α are the regression density = 20.59 kN/m3; stress levels and reduced stress
parameters. - Wopt = 5.5%. levels.
QS3Sandy gravel GWWell-graded Rahman and q - q is deviatoric stress; - Gs = 2.63;
(GrSa)Cu 20; Cc 1.5 sandy gravel Erlingsson bS pa - p is mean stress; - Fines Content = 2.2%;
p (N ) = N f Sf Sf =
[66] p - pa atmospheric pressure; - Maximum dry
pa - a, b and α are the regression density = 20.35 kN/m3;
parameters. - Wopt = 5.5%.
QS3 GWWell-graded Gu et al. 2sin c 6cos - J2 is the second invariant of the - γd = 2.162 kg/m3; - This model used the shear
p = (N) ( K =
Grave(granite [35] 0e J2 )m ( I1 + K )n =
3 (3 sin ) 3 (3 sin ) deviatoric stress tensor; - ω = 6.7%; strength ratio and stress terms
aggregate) - I1 is the first invariant of the - PI = 4; LL = 25; in the Drucker-Prager model to
stress tensor; - ϕ = 51.3°; c = 20.2 kPa; construct a stress-dependent
- c and ϕ are the cohesion and - Cu≈25; Cc≈2.8. model;
friction angle, respectively; - The stress term J2 represents the
- ε0, ρ, β, m and n are model softing effects that deviatoric
Transportation Geotechnics 23 (2020) 100326

coefficients. shear stress has on the material,

(continued on next page)
A. Ramos, et al. Transportation Geotechnics 23 (2020) 100326

- A higher value of I1 + K leads to

number. The authors carried out a

and the term I1 + K indicates the

friction angle, and also α and K by

strength parameter (cohesion and
Some models were developed using certain SI units, and where

UGM; it is highly affected by the

plotting the diagram J2 – I1and

smaller plastic deformation. The
- The power m is always positive
and a larger value yields larger

hardening/strengthening effect
possible these units are described in the column “variables and empirical

that hydrostatic stress has on

power n is always a negative

Repeated Load Triaxial Tests

compressive strength test to
determine the cohesion, the
constants”.
permanent deformation; The permanent deformation is a dimensionless parameter that can

(N = 10,000 cycles).
also be referred as permanent strain. In fact, the nomenclature varies, as
mentioned above. The permanent strain is defined as change in the

friction angle);
length of the material and expressed as a function of the length being
Observations

changed. The plastic/permanent deformation is defined as not re-

coverable. The materials that are part of the pavements and railway
structures accumulate some amount of permanent deformation due to
the traffic cyclic load. The stress parameters (induced by the cyclic
- PI = non-plastic; LL = not
Parameters and characteristics

loads) are one of the main inputs of these models. The inclusion of these
- ϕ = 54.9°; c = 66.2 kPa;

parameters in the permanent deformation model varies as well as the

- γd = 1.934 kg/m3;

nomenclature. Indeed, there are models dependent on the p and q

- Cu≈45; Cc≈1.25.

(mean and deviatoric) stresses or on the octahedral stresses, for ex-

- ω = 13.5%;

ample.
applicable;

In the permanent deformation tables, the deviator stress is re-

presented by qd, q or σd, the variable for the atmospheric pressure/re-
ference stress is represented by σatm, p0, pa or pref, while the equation
models were developed under different conditions, so permanent de-
- J2 is the second invariant of the

formation is represented in the following tables by the nomenclatures:

Variables and Empirical Constants

- c and ϕ are the cohesion and

- I1 is the first invariant of the

friction angle, respectively;

- ε0, ρ, β, m and n are model

εp, εp1,N, zp , εq;r (axial residual strain), 1p , εp (N), 1p (N ) . The constants of

deviatoric stress tensor;

the models used to fit the model to the available data are represented by
different letters and symbols which are defined as material parameters,
model constants, and model parameters and soil parameters, as listed in
stress tensor;

the permanent deformation tables.

coefficients.

References
3 (3 sin )

[1] Abdelkrim M, Bonnet G, Buhan PD. A computational procedure for predicting the
c 6cos

long term residual settlement of a platform induced by repeated traffic loading.

Comput Geotech 2003;30:463–76.
[2] Al Shaer A, Duhamel D, Sab K, Foret G, Schmitt L. Experimental settlement and
=

dynamic behavior of a portion of ballasted railway track under high speed trains. J
K
3 (3 sin )

Sound Vibration 2008;316:213–33.

[3] Alves Costa P, Lopes P, Silva Cardoso A. Soil shakedown analysis of slab railway
2sin

tracks: Numerical approach and parametric study. Transp Geotech 2018;16:85–96.

[4] Ara I, Division EC. Guide for mechanistic-empirical design of new and rehabilitated
pavement structures (Final report, NCHRP project 1–37A). Washington, DC:
J2 )m ( I1 + K )n =

Transportation Research Board of the National Academies; 2004.

[5] Ba M. Correlation between resilient modulus and permanent deformation during a
large scale model experiment of unbound base course. Lecture Notes Civil Eng 2018.
[6] Barksdale RD. Laboratory evaluation of rutting in base course materials. Proc. 3rd
Int. Conf on the Structural Design of Asphalt Pavements London. 1972.
[7] Bian X, Jiang H, Cheng C, Chen Y, Chen R, Jiang J. Full-scale model testing on a
Equation model

ballastless high-speed railway under simulated train moving loads. Soil Dyn
(N) (

Earthquake Eng 2014;66:368–84.

[8] Brown S, Yu H, Juspi HJW. Validation experiments for lower-bound shakedown
0e

theory applied to layered pavement systems. Géotechnique 2012;62:923–32.

=

[9] Brown SF, Hyde AF. Significance of cycling confining stress in repeated-load triaxial
p

testing of granular material. Transp. Res. Rec. Washington D.C: Transportation

Research Board; 1975. p. 49-–58.
[10] Cai Y, Sun Q, Guo L, Juang CH, Wang J. Permanent deformation characteristics of
Gu et al.
Author’s

saturated sand under cyclic loading. Can Geotech J 2015;52:795–807.

model

[35]

[11] Chai J-C, Miura N. Traffic-load-induced permanent deformation of road on soft soil.
J Geotech Geoenviron Eng 2002;128:907–16.
[12] Chan AH. Permanent deformation resistance of granular layers in pavements PhD
thesis England: University of Nottingham; 1990.
Gravel(limestone
GWWell-graded

[13] Chazallon, C. 2000. An elastoplastic model with kinematic hardening for unbound
conglomerate
classification

aggregates in roads. UNBAR 5 Conference. Nottingham.

aggregate)

[14] Chazallon C, Allou F, Hornych P, Mouhoubi S. Finite elements modelling of the

long-term behaviour of a full-scale flexible pavement with the shakedown theory.
ASTM

Int J Numer Anal Meth Geomech 2009;33:45–70.

[15] Chazallon C, Habiballah T, Hornych P. Elastoplasticity framework for incremental
or simplified methods for unbound granular materials for roads. Lisbon: BCRA
workshop on modelling of flexible pavements; 2002.
[16] Chazallon C, Hornych P, Mouhoubi S. Elastoplastic model for the long-term beha-
vior modeling of unbound granular materials in flexible pavements. Int J Geomech
Table A4 (continued)

2006:279–89.
[17] Chazallon C, Koval G, Hornych P, Allou F, Mouhoubi S. Modelling of rutting of two
flexible pavements with the shakedown theory and the finite element method.
Comput Geotech 2009;36:798–809.
[18] Chazallon C, Koval G, Mouhoubi S. A two-mechanism elastoplastic model for sha-
kedown of unbound granular materials and DEM simulations. Int J Numer Anal
QS3
UIC

Meth Geomech 2012;36:1847–68.

24
A. Ramos, et al. Transportation Geotechnics 23 (2020) 100326

[19] Chen R, Chena J, Zhaob X, Bian X, Chen Y. Cumulative settlement of track subgrade Geotech Geoenviron Eng 1996;122:1006–13.
in high-speed railway under varying water levels. Int J Rail Transportat [50] Ling X, Li P, Zhang F, Zhao Y, Li Y, An L. Permanent deformation characteristics of
2014;2:205–20. coarse grained subgrade soils under train-induced repeated load. Adv Mater Sci Eng
[20] Chow LC, Mishra D, Tutumluer E. Framework for development of an improved 2017;2017:15.
unbound aggregate base rutting model for mechanistic-empirical pavement design. [51] Liu S, Wang J. Application of shakedown theory in track substructure design.
Transport Res Record J Transport Res Board 2014;2401:11–21. Proceedings of the Institution of Civil Engineers – Ground Improvement. 2019.
[21] Collins I, Boulbibane M. Geomechanical analysis of unbound pavements based on [52] Lytton RL, Uzan J, Fernando EG, Roque R, Hiltunen D, Stoffels SM. Development
shakedown theory. J Geotech Geoenviron Eng 2000;126:50–9. and validation of performance prediction models and specifications for asphalt
[22] Coronado O, Caicedo B, Taibi S, Correia AG, Fleureau J-M. A macro geomechanical binders and paving mixes. Washington, D.C: Report SHRPA-357: TRB, National
approach to rank non-standard unbound granular materials for pavements. Eng Research Council; 1993.
Geol 2011;119:64–73. [53] Momoya Y, Sekine EFT. Deformation characteristics of railway roadbed and sub-
[23] Coronado O, Caicedo B, Taibi S, Correia AG, Souli H, Fleureau J-M. Effect of water grade under moving-wheel load. Soils Found 2005;45:99–118.
content on the resilient behavior of non standard unbound granular materials. [54] Monismith CL, Ogawa N, Freeme CR. Permanent deformation characteristics of
Transp Geotech 2016;7:29–39. subgrade soils due to repeated loading. Transp Res Rec 1975;537:1–17.
[24] Correia AG. Innovations in design and construction of granular pavements and [55] Orobio A, Zaniewski JP. Sampling-based sensitivity analysis of the mechanistic-
railways. Advances in Transportation Geotechnics -. Proceedings of the 1st empirical pavement design guide applied to material inputs. J. Trans. Res. Rec.
International Conference on Transportation Geotechnics. 2008. p. 3–13. 2011;2226:95–193.
[25] Desai CS. A general basis for yield, failure and potential function in plasticity. Int. J. [56] Pappin JW. Characteristics of a granular material for pavement analysis. University
Numer. Anal. Methods Geomech. 1980;4:361–75. de Nottingham; 1979. PhD Thesis.
[26] Desai CS, Faruque MO. Constitutive model for geologic materials. J. Eng. Mech. Div. [57] Paute J, Hornych P, Benaben J. Comportement mecanique des graves non traitees.
1984:1391–408. Bull Liasion Laboratories Ponts Chaussées 1988;156:21–36.
[27] En13286-7 2004a. Unbound and hydraulically bound mixtures. Part 7: Cyclic load [58] Paute JL, Hornych P, Benaben JP. Comportement mécanique des graves non
triaxial test for unbound mixtures Brussels.: CEN-European Committee for traitées. Bulletin de Liaison des Laboratoires des Ponts et Chaussées
Standardization. 1994;190:27–38.
[28] Erlingsson S, Rahman MS. Evaluation of permanent deformation characteristics of [59] Puppala AJ, Mohammad LN, Allen A. Permanent deformation characterization of
unbound granular materials by means of multistage repeated-load triaxial tests. subgrade soils from RLT test. J Mater Civ Eng 1999;11:274–82.
Transport Res Rec J Transport Res Board 2013;2369:11–9. [60] Puppala AJ, Saride S, Chomtid S. Experimental and modeling studies of permanent
[29] Erlingsson S, Rahman S, Salour F. Characteristic of unbound granular materials and strains of subgrade soils. J Geotech Geoenviron Eng 2009;135:1379–89.
subgrades based on multi stage RLT testing. Transp Geotech 2017;13:28–42. [61] Qamhia IIA, Moaveni M, Byun Y-H, Feng B, Tutumluer E. Implementation frame-
[30] Gidel G, Hornych P, Chauvin J-J, Breysse D, Denis A. A new approach for in- work of the UIUC aggregate base rutting model. Int J Pavement Eng 2019:1–13.
vestigating the permanent deformation behaviour of unbound granular material [62] Qian J, Lin H, Gu X, Xue J. Dynamic shakedown limits for flexible pavement with
using the repeated load triaxial apparatus. Bulletin des Laboratoires des Pont et cross-anisotropic materials. Road Mater Pavement Des 2018:1–21.
Chaussées 2001;233:5–21. [63] Qian J, Li Shiyuan, Xiaoqiang Gu, Zhang J. A unified model for estimating the
[31] Gomes Correia A. Influence of compaction conditions on resilient and permanent permanent deformation of sand under a large number of cyclic loads. Ocean Eng
deformations of aggregates mixtures of granite. Gomes Correia and Quibel 2019;181:293–302.
(Editors), Compaction of soils and granular materials. Paris: Presse Nationale des [64] Qian J, Wang Y, Wang J, Huang M. The influence of traffic moving speed on sha-
Ports et des Chaussées, 2000. kedown limits of flexible pavements. Int J Pavement Eng 2019;20:233–44.
[32] Gomes Correia, A. 2004. Evaluation of mechanical properties of unbound granular [65] Rahman MM, Gassman SL. Permanent deformation characteristics of coarse grained
materials for pavements and rail tracks. Gomes Correia and Loizos (Editors), subgrade soils using repeated load triaxial tests. Philadelphia, Pennsylvania: Geo-
Geotechnics in Pavement and Railway Design and Construction., MillPress. Congress 2019; 2019.
[33] Grabe PJ, Clayton CR. Permanent deformation of railway foundations under heavy [66] Rahman MS, Erlingsson S. A model for predicting permanent deformation of un-
axle loading. In: Conference, P. O. I. H. H., ed. Conference of the International bound granular materials. Road Mater Pavement Des 2015;16:653–73.
Heavy Haul Association, International Heavy Haul Association, 2003. 325–333. [67] Salour F, Erlingsson S. Characterization of permanent deformation of silty sand
[34] Grégoire C, Correia AG, De Bel R, Dethy B. Mechanical behavior of natural and subgrades from multistage RLT tests. The 3rd International Conference on
recycled granular materials for roads. J Test Evaluat 2011;39. Transportation Geotechnics (ICTG 2016) Guimarães, Portugal. 2016. p. 300–7.
[35] Gu F, Zhang Y, Droddy CV, Luo R, Lytton RL. Development of a new mechanistic [68] Sun Q, Indraratna B, Ngo NT. Effect of increase in load and frequency on the re-
empirical rutting model for unbound granular material. J. Mater. Civ. 2016;28. silience of railway ballast. Géotechnique 2019;69:833–40.
[36] Guo, L. Experimental study on the static and cyclic behavior of saturated soft clay [69] Sweere GTH. Unbound granular bases for roads. Delft University of Technology,
under complex stress path. PhD Thesis, Zhenjiang Univ., Hangzhou, China (in Netherlands; 1990. PhD Thesis.
Chinese), 2013. [70] Tseng KH, Lytton RL. Prediction of permanent deformation in flexible pavement
[37] Hornych P, Chazallon C, Allou F, El Abd A. Selection and evaluation of models for materials. In: Schrauders HG, Marek CR, editors. Implication of aggregates in de-
prediction of permanent deformations of unbound granular materials in road pa- sign, construction, and performance of flexible pavements. Philadelphia. PA:
vements. Work Package 5, Performance-based specifications. SAMARIS; 2004. American Society for Testing and Materials; 1989. p. 154–72.
[38] Huurman M. Permanent deformation in concrete block pavement. Delft University [71] Tseng KH, Lytton RL. Prediction of permanent deformation in flexible pavements
of Technology, Netherlands; 1997. PhD Thesis. materials, implication of aggregates in the design, construction, and performance of
[39] Hyde AFL. Repeated load triaxial testing of soils. Nottingham: University of flexible pavements. ASTM STP 1016, ASTM, West Conshohocken PA. 1989. p.
Nottingham; 1974. PhD thesis. 154–72.
[40] Indraratna B, Babar Sajjad M, Ngo T, Gomes Correia A, Kelly R. Improved perfor- [72] Ullidtz P. Mathematical model of pavement performance under moving wheel load.
mance of ballasted tracks at transition zones: a review of experimental and mod- J. of Trans. Res. Rec., Washington, D.C. 1993;1384:94–9.
elling approaches. Transp Geotech 2019:21. [73] Uzan J. Permanent deformation of a granular base material. Transp. Res. Rec.
[41] Indraratna B, Ngo NT, Rujikiatkamjorn C. Deformation of coal fouled ballast sta- 1999;1673:89–94.
bilized with geogrid under cyclic load. J Geotech Geoenviron Eng [74] Vermeer, P. A. A five-constant model unifying well-established concepts. Proc., Int.
2013;139:1275–89. Workshop on Constitutive Relations for Soils, 1982. A.A. Balkema, Rotterdam,
[42] Jouve P, Martinez J, Paute JL, Ragneau E. Rational model for the flexible pave- Netherlands, 175–197.
ments deformations. Proceedings of the 6th International Conference on Structural [75] Wang J, Yu H. Residual stresses and shakedown in cohesive-frictional half-space
Design of Asphalt pavements. Ann Arbor; 1987. p. 50–64. under moving surface loads. Geomech Geoeng 2013;8:1–14.
[43] Khedr S. Deformation characteristics of granular base course in flexible pavement. [76] Wei X, Wang G, Wu R. Prediction of traffic loading–induced settlement of low-
Transp Res Rec 1985;1043:131–8. embankment road on soft subsoil. Int J Geomech 2017:17.
[44] Korkiala-Tanttu L. A new material model for permanent deformations in pave- [77] Werkmeister S. Permanent deformation behavior of unbound granular materials
ments. Proc. of the Seventh Conference on Bearing Capacity of Roads and Airfields PhD Thesis Dresden, Germany: University of Technology; 2003.
Trondheim, Norway. 2005. p. 10. [78] Werkmeister S, Dawson A, Wellner F. Permanent deformation behaviour of gran-
[45] Lekarp F. Resilient and permanent deformation behavior of unbound aggregates ular materials. Road Mater. Pavement Des. 2005;6:31–51.
under repeated loading. The Royal Institute of Technology, KTH, Stockholm, [79] Werkmeister S, Numrich R, Wellner F. The development of a permanent deforma-
Sweden; 1999. tion design model for unbound granular materials with the shakedown-concept.
[46] Lekarp F, Dawson A. Modelling permanent deformation behaviour of unbound Proc. of the 6th International Conference on the Bearing Capacity of Roads and
granular materials. Constr Build Mater 1998;12:9–18. Airfields Lisbon, Portugal. 2002. p. 1081–96.
[47] Lekarp F, Isacsson U, Dawson A. Permanent strain response of unbound aggregates. [80] Wolff H, Visser AT. Incorporating elasto-plasticity in granular layer pavement de-
J Transp Eng 2000;126:76–83. sign. Proceedings of the Institution of Civil Engineers-Transport. 1994. p. 259–72.
[48] Lentz RW, Baladi GY. Simplified procedure to characterize permanent strain in sand [81] Xiao J, Juang C, Wei K, Xu S. Effects of principal stress rotation on the cumulative
subjected to cyclic loading. Proceedings Int. Symposium on Soils Under Cyclic and deformation of normally consolidated soft clay under subway traffic loading. J
Transient Loading Swansea. 1980. p. 89–95. Geotech Geoenviron Eng 2014:140.
[49] Li D, Selig ET. Cumulative plastic deformation for fine-grained subgrade soils. J [82] Xiao Y, Tutumluer E, Mishra D. Performance evaluation of unbound aggregate

25
A. Ramos, et al. Transportation Geotechnics 23 (2020) 100326

permanent deformation models for different aggregate physical properties. [84] Yang S-R, Huang W-H, Liao C-C. Correlation between resilient modulus and plastic
Transport Res Record J Transport Res Board 2015;2525:20–30. deformation for cohesive subgrade soil under repeated loading. Transport Res
[83] Yang S-R, Huang W-H. Permanent deformation and critical stress of cohesive soil Record J Transport Res Board 2008:2053.
under repeated loading. Transport Res Record J Transport Res Board [85] Yesufa GY, Hoffa I. Finite element modelling for prediction of permanent strains in
2007;2016:23–30. fine-grained subgrade soils. Road Mater Pavement Des 2015;16:392–404.

26