You are on page 1of 38

Journal Pre-proofs

Evaluation of Flexible Pavement Deterioration Conditions Using Deflection


Profiles Under Moving Loads

Yong Deng, Xue Luo, Yazhou Zhang, Robert L. Lytton

PII: S2214-3912(20)30322-6
DOI: https://doi.org/10.1016/j.trgeo.2020.100434
Reference: TRGEO 100434

To appear in: Transportation Geotechnics

Received Date: 9 March 2020


Revised Date: 6 July 2020
Accepted Date: 15 August 2020

Please cite this article as: Y. Deng, X. Luo, Y. Zhang, R.L. Lytton, Evaluation of Flexible Pavement
Deterioration Conditions Using Deflection Profiles Under Moving Loads, Transportation Geotechnics (2020),
doi: https://doi.org/10.1016/j.trgeo.2020.100434

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover
page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version
will undergo additional copyediting, typesetting and review before it is published in its final form, but we are
providing this version to give early visibility of the article. Please note that, during the production process, errors
may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

© 2020 Elsevier Ltd. All rights reserved.


Evaluation of Flexible Pavement Deterioration Conditions Using Deflection
Profiles Under Moving Loads

Yong Deng
Zachry Department of Civil and Environmental Engineering
Texas A&M University
3136 TAMU, DLEB 508K, College Station, Texas, 77843
Email: dengyong1992@tamu.edu

Xue Luo, Ph.D.


College of Civil Engineering and Architecture
Zhejiang University
866 Yuhangtang Road, An-zhong Bldg., Hangzhou, China, 310058
Email: xueluo@zju.edu.cn

Yazhou Zhang
CCCC Second Highway Consultants Co., Ltd
Wuhan 430052, Hubei, China
Email: yazhouzhang@whut.edu.cn

Robert L. Lytton, Ph.D., P.E.


Zachry Department of Civil and Environmental Engineering
Texas A&M University
3136 TAMU, DLEB 503A, College Station, Texas 77843
Email: r-lytton@civil.tamu.edu
2

Abstract
With the development and application of highway-speed nondestructive testing (NDT) devices, a
reliable method to evaluate the current level of deterioration of pavements from the measured
results is necessary to provide comprehensive and timely maintenance and rehabilitation
decisions for pavements. The objective of this study is to evaluate the deterioration conditions of
flexible pavements from their deflection profiles under the moving loads. The deterioration is
due primarily to the development and growth of microcracks in flexible pavement materials and
is commonly called the “crack initiation” phase. This phase is characterized and represented by
changes in the dynamic modulus and the phase angle. To simulate a flexible pavement under
moving loads, a three-dimensional (3D) finite element (FE) model and its equivalent two-
dimensional (2D) axisymmetric FE model are constructed using a commercial FE software and
the technique of artificial intelligence (AI). The deflection profiles of the pavement model are
analyzed and the time lag between the load and deflection peaks is used to define a new term
named “lag angle” to represent the structural response of the flexible pavement under the moving
loads. It is also found the lag angle is closely related to the degree of deterioration of the
pavement, the speed of the moving load, the structural and material properties of the pavement,
which reveals a promising application of the lag angle in the evaluation of flexible pavements
using highway-speed NDT devices.

Keywords: nondestructive testing; pavement deterioration condition; finite element updating;


surrogate model; Artificial Intelligence
3

1 Introduction
An instant and accurate evaluation of the current level of microcracking in flexible pavements is
of essential value for the efficient scheduling of pavement maintenance and rehabilitation. In the
area of the nondestructive testing (NDT), this goal is being approached by the development of
the highway-speed NDT devices. Compared with the traditional NDT devices such as the Falling
Weight Deflectometer (FWD), the highway-speed devices have the advantage that no traffic
control is required during the test, which ensures the safety to the technicians. In addition, the
highway-speed NDT devices significantly enhance the productivity of the test, which results in a
timely evaluation and decision-making for pavements (Deng et al., 2019).
The widely used highway-speed NDT devices are the Rolling Wheel Deflectometer
(RWD) and the Traffic Speed Deflectometer (TSD). The RWD is manufactured by Applied
Research Associates, Inc to measure the pavement surface deflections. It applies a one-half 18-
kip single axle load and travels at a highway speed up to 105 km/h (Diefenderfer, 2010; Deng et
al., 2019). The distances between the lasers and the pavement surface when it is deflected and
undeflected are recorded to calculate the pavement deflections. The TSD is another highway-
speed device developed by Greenwood Engineering, Inc. It applies a 100 kN wheel load which
can move at a speed up to 96 km/h. The TSD applies the Doppler-based laser technology to
measure the vertical deflection velocity of the pavement, which can be used to calculate the
deflection and the deflection slope of the pavement (Golding et al., 2017; Deng et al., 2019).
The operational speeds of the RWD and TSD ideally simulate the highway speeds of
traffic loads so that the information collected from RWD and TSD represents pavements under
in-service conditions. In the current pavement condition evaluations, RWD is mainly used to
obtain the structural indices of pavements from measured deflections. The structural number
(SN) and structural condition index (SCI) calculated using the measured deflections at the
loading center show good correlations with those obtained using FWD (Elbagalati et al., 2016;
Elbagalati et al., 2017). And these two indices can be incorporated into the decision-making
system for pavements (Elbagalati et al., 2017). For TSD, in addition to the structural indices such
as the loading center deflection (Chai et al., 2016; Golding et al., 2017), the adjusted structural
number (SNP) (Golding et al., 2017) and the deflection slope index (DSI) (Nasimifar et al.,
2016; Nasimifar et al., 2018), the deflections and deflection slopes obtained from TSD can be
used to backcalculate the layer moduli (Nasimifar et al., 2017; Nasimifar et al., 2018) and
4

predict responses at critical locations (Nasimifar et al., 2016; Nasimifar et al., 2017; Nasimifar et
al., 2018). The information can then be used to evaluate the pavement deteriorated conditions
and estimate their remaining service life.
Because continuous deflection profiles of pavements under moving loads can be provided
by highway-speed NDT devices, a new method to evaluate the flexible pavement condition from
its deflection profile under a moving load is proposed. Advanced techniques such as the finite
element method (FEM) and artificial intelligence (AI) were applied in developing this method. In
the next section, the methods to evaluate the deterioration conditions of the flexible pavement
materials are introduced. Then, a three-dimensional (3D) FE model and its equivalent two-
dimensional (2D) axisymmetric FE model are constructed for the simulation of a deteriorating
flexible pavement under moving loads. Following that, the deflection profiles of the pavement
are analyzed for a relationship between the changing material properties, pavement responses
and the speed of the moving load. Finally, the major findings and contributions are summarized.

2 Fatigue Damage Evaluation of Asphalt Mixtures


Fatigue is a major distress and failure mode of pavement caused by the load repetitions and
climatic effects. It results in the degradation of the pavement materials and structures which
significantly influences the riding quality and safety, and reduces the service life of pavements
(Cocurullo et al., 2008; Moreno-Navarro and Rubio-Gámez, 2016). A comprehensive
understanding of the fatigue mechanism of the materials and a precise fatigue criterion can play
an essential role in the fatigue behavior evaluation and fatigue life prediction, and eventually
contribute to the pavement design, maintenance and rehabilitation (Kim et al., 2003). Over the
years, analysis has been conducted using numerous fatigue tests (Di Benedetto et al., 2003; Kim
et al., 2003; Cocurullo et al., 2008; Tapsoba et al., 2012; Boudabbous et al., 2013; Ahmed and
Khalid, 2015; Poulikakos et al., 2015; Tapsoba et al., 2015; Qabur, 2018) to evaluate the fatigue
performance of the asphalt mixtures, which are the materials of flexible pavement surfaces.
As shown in Figure 1, by analyzing the modulus and the phase angle of asphalt mixtures
with load cycles in the fatigue tests, three phases were identified based on the curve shapes and
processes within the materials (Tapsoba et al., 2012; Di Benedetto et al., 2003; Poulikakos et al.,
2015; Tapsoba et al., 2015; Moreno-Navarro and Rubio-Gámez, 2016). In Phase I, the modulus
undergoes a rapid decrease and the phase angle undergoes a rapid increase, which are caused by
5

the heating and thixotropy due to the initial loading process (Di Benedetto et al., 2011). In Phase
II, the decrease in the modulus and the increase in the phase angle are quasi-linear with load
cycles, which result from the initiation and propagation of micro-cracks. Following that, the
coalescence of the micro-cracks creates macro-cracks, which propagate in the Phase III. Both the
modulus and the phase angle begin to drop quickly when entering Phase III.

Figure 1. Normalized Modulus and Phase Angle with Load Cycles in a Fatigue Test

From the test results, four types of criteria are widely used to evaluate the material
condition (Tapsoba et al., 2012) in laboratory fatigue tests:
1. Modulus/Stiffness
The values of modulus and stiffness decrease continuously with load repetitions in
fatigue tests. A 50% or 30% decrease in the modulus or stiffness from its initial value is assumed
to be the failure of the material (Kim et al., 2003; Tapsoba et al., 2012).
2. Phase Angle
Since the phase angle has its peak value before the macrocracks occur in the material. It
can serve as an indicator for the fatigue damage degree of the material. Reese (Reese, 1997)
proposed this criterion based on the phase angle evolution.
6

3. Homogeneity
With the occurrence of macro-cracks, the material is not homogeneous anymore
(Tapsoba et al., 2012). Accordingly, the material responses measured at different locations of the
sample vary significantly. The strain amplitude and phase angle were used as the indicator
(Tapsoba et al., 2012).
4. Dissipated Energy
The dissipated energy is the energy done by the external work, which is calculated from
the area under the strain-stress curve in the fatigue test (Shen et al., 2006; Luo et al., 2020). The
criterion for the fatigue failure of the material were defined based on the change in the evolution
of the dissipated energy (Pronk and Hopman, 1991) or the dissipated energy ratio (Carpenter et
al., 2003) with load cycles.
This paper proposes to extend the substance of these criteria to pavements in the field.
The modulus and the phase angle are two essential properties characterizing the asphalt mixtures,
the similar trends of the modulus and the phase angle were implemented into the pavement
model to reflect the deterioration of the pavement material. The responses of the pavement were
evaluated for all potential relationships between the structural responses and material properties.
Once the relationship can be determined, the structural response will serve as a rapid indicator to
reflect current deterioration conditions of pavements and provide an advance warning of the
pavement failure.

3 Model Updating Algorithm for Equivalence between 3D and 2D Axisymmetric FE


Models
To represent pavement structures under moving loads, 3D FE models have the advantages of
accurate model geometries and loading configurations over 2D axisymmetric models. In
addition, because the 2D axisymmetric models are essentially cylinders, neither non-circular nor
moving loads can be applied in the 2D axisymmetric models due to model assumptions.
However, compared with the 3D FE models, the 2D axisymmetric models require significantly
less computation time and storage space. Therefore, in the pavement design and analysis, the 2D
axisymmetric models are also widely applied.
In order to simulate moving loads in 2D axisymmetric models, research has been
conducted to build the relationship between the moving loads and dynamic loads. The most
7

commonly used equivalent method is that the cycle of the dynamic load is the division of the
distance to the speed of the moving load (Huang 1993; Wang et al., 2017). The distance refers to
the length of the contact area between the tires and the pavement surface or the length of the
structure which the moving load passes through. Alternatively, the dynamic loads can be
equivalent to the moving loads by comparing the pavement responses under two types of loads,
in which the pulse duration is typically used. The mechanistic-empirical design guide (MEPDG)
(NCHRP, 2004) and Al-Qadi et al. (Al-Qadi et al., 2008) applied the pulse duration of the
vertical stress within the asphalt concrete layer to determine the equivalent loading frequency.
Garcia and Thompson (Garcia and Thompson, 2008) studied the pulse duration of the tensile
strain in transverse and longitudinal directions of the pavement under moving loads. Wang and
Li (Wang and Li, 2016) obtained the equivalent moving speeds to the FWD loading based on the
pulse durations of the compressive stress, tensile and shear strains. Such an equivalent method
has one limitation. The pulse duration varies with the surface layer depth and the type of
pavement responses. The determined equivalent loading frequencies or speeds should be checked
with the material or pavement performance to meet the design and analysis requirements (Ulloa
et al., 2013).
Based on the objective of this study which is determining the pavement deterioration
conditions from the moving pavement surface deflections, a more targeted equivalent method is
proposed to build a relationship between the moving loads and the dynamic loads.
The determination of the equivalent dynamic load to the moving load requires a model
updating method based on artificial intelligence (AI). Artificial intelligence is the technology
developing computer programs to solve complicated tasks in the way of human intelligence (Lu
et al., 2012; Krishnamoorthy and Rajeev, 2018). Artificial intelligence covers a wide range of
techniques and applications. In the field of civil engineering, the artificial neural network (ANN)
method is typically used in finding correlations between parameters and predicting material and
structural behaviors (Ling et al., 2017; Gu et al., 2018; Saha et al., 2018). In the structural health
monitoring (SHM), the particle swarm optimization (PSO) method is used to minimize the
difference between the design properties and the real structures (Qin et al., 2018b). In this study,
a Kriging model and the PSO algorithm were applied to the 2D axisymmetric model updating in
terms of the load magnitude and cycle of the dynamic load to obtain the same required responses
8

as in 3D models applied with the moving load. The concepts involved and the whole procedures
are presented below.
3.1 3D FE simulation of deflection under moving loads
In order to obtain the responses of a flexible pavement under moving loads, a 3D finite element
(FE) model was built using the software ABAQUS (ABAQUS, 2012). Figure 2(a) and Table 1
show the structural configuration of the model and the basic properties of the layer materials.
The pavement is a three-layer structure consisting of a viscoelastic surface, an elastic base and an
elastic subgrade. The Prony series coefficients listed in Table 1 were applied in a representative
analytical model (Schapery 1961) describing the relaxation modulus of the pavement surface
made of asphalt concrete (Gu et al., 2016). The Prony series model of the relaxation modulus is
expressed in Equation 1 (ABAQUS, 2012).
N
E (t )  E    E i e  t / i (1)
i 1

where E  =long-term modulus;


E i =Prony regression coefficients;
 i =relaxation time;
N =number of Prony series terms.

Model
Value
Parameters
Surface Layer
0.10
Thickness (m)
Base Layer
Thickness (m) 0.15
Subgrade
Thickness (m) 1.78
Horizontal
Length (m) 13.72
Transverse
Length (m) 5.49
9

(a) The 3D Pavement Structure and Parameters Used in the FE Analysis

0.06 m

0.21 m

0.18 m

(b) Tire-Pavement Contact Stresses

(c) Moving of Contact Stresses


Figure 2. The 3D Pavement FE Model with the Moving Load

Table 1. The Materials Properties Used in the FE Analysis


Prony Series Coefficients
i E i , MPa i , s
1 6564.22 4.06E-6
2 6582.36 2.56E-4
Surface 3 3200.51 7.71E-3
Layer 4 1341.86 2.10E-1
5 299.20 3.88E+0
6 103.36 6.53E+1
 41.71
Density=2400 kg/m3, Poisson’s Ratio=0.35, Damping Ratio=0.05
10

Modulus=250 MPa, Density=1800 kg/m3, Poisson’s Ratio=0.35,


Base Layer
Damping Ratio=0.05
Modulus=100 MPa, Density=1600 kg/m3, Poisson’s Ratio=0.40,
Subgrade
Damping Ratio=0.05

The load of a typical dual-tire vehicle is simulated by two loading paths symmetrically
created along the horizontal direction and shown in Figure 2(a). The details can be seen in the
magnified loading paths in Figure 2(b). The 0.06-m space between two loading paths represents
the distance between two tire edges and the 0.21-m width of each loading path represents the
width of the contact area between the tire and the pavement surface. For simplicity, the contact
stress of each tire is assumed to be a constant pressure of 0.7 MPa uniformly distributed on the
contact area (Al-Qadi et al., 2002). As shown in Figure 2(c), to simulate the movement of the
vehicle, the contact area was divided into three equal elements in the horizontal direction and the
contact area was moved forward by one element at a time in successive analysis steps in
ABAQUS. According to different moving speeds of the vehicle, the duration of each analysis
step was calculated by dividing the length of one element by the moving speed.
The movement of the vehicle along the pavement surface was simulated and the vertical
deflections of the midpoint at the pavement surface were collected. Figure 3(a) shows the
locations of the points where the vehicle started to move and the vertical deflections were
collected. The collected vertical deflections of the midpoint with time when the moving speed is
80.47 km/h (50 mph) are presented in Figure 2(a). In addition to the deflections, the vertical
deflection history includes the response time and duration which can be seen as characteristics of
the pavement system.
Since the results obtained in ABAQUS are discrete, a Fourier Series function expressed
in Equation 2 was applied to fit the deflection history for better capturing the characteristics of
the curve such as the end points shown in Figure 3(a). The maximum vertical deflection and the
time to reach the maximum vertical deflection are also shown in Figure 3(a).
N
D(t )  a 0   [a i cos(it )  bi sin(it )] (2)
i 1

N
D(t )   [i a i sin(it )  ibi cos(it )] (3)
i 1

where D(t ) , D (t ) =vertical deflection and its first derivative with respect to time;
11

N =number of Fourier series terms;


a 0 , a i , bi ,  = Fourier series coefficients.

Figure 3(a) shows a desirable match between the results obtained in ABAQUS and
predicted by the Fourier Series function, which indicates the accuracy of this analytical function.
The location of the end points can be determined directly by setting Equation 2 to zero, and the
location and the value of maximum vertical deflection can be calculated by setting Equation 3 to
zero then substituting the obtained time into Equation 2.

(a) The Vertical Deflection History Predicted by Fourier Series Function under Moving
Load of 80.47 km/h
12

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4


0
Time, s
Vertical Deflection, mm

-0.1

40.23 km/h
-0.2
56.33 km/h
72.42 km/h
-0.3 80.47 km/h
96.56 km/h
112.65 km/h
-0.4 128.75 km/h
160.93 km/h

-0.5
(b) The Vertical Deflection History under Moving Load of Different Speeds
Figure 3. The Vertical Deflection History under Moving Loads

In order to compare the pavement responses under the moving loads of different speeds,
the fitted deflection history curve was adjusted with its origin at its left end, which represents the
moment when the pavement starts to deflect due to the approaching load. Figure 3(b) shows the
vertical deflection histories of the same pavement under moving loads at different speeds. It can
be seen that with the increase of the moving speed, the maximum vertical deflection decreases
and the response duration decreases. Table 2 presents the maximum deflections and the times to
reach the maximum deflections in the cases of different moving loads calculated from Equation 2
and 3.

Table 2. The 3D FE Analysis Results


Speed, km/h Max Deflection, mm Time to Reach the Maximum Deflection, s
40.23 4.90E-1 1.76E-1
56.33 4.72E-1 1.26E-1
72.42 4.60E-1 9.75E-2
80.47 4.52E-1 8.77E-2
96.56 4.39E-1 7.21E-2
112.65 4.27E-1 6.19E-2
13

128.75 4.17E-1 5.34E-2


160.93 3.96E-1 4.12E-2

3.2 2D Axisymmetric FE Simulation of Deflection under Dynamic Loads


The 2D axisymmetric model was built in ABAQUS with the same structural configuration and
material properties as those in the 3D FE model. Figure 4(a) shows the geometry of the 2D
axisymmetric model and the applied dynamic load with the load magnitude F and the load cycle
T. The 0.15-m radius of the loading area is identical to the radius of the standard loading plate of
the FWD, in which the dynamic load is uniformly distributed. For consistency, the loading area
is fixed in this study so that the dynamic load in the following text is also represented by the
pressure magnitude P and the cycle T.

(a) The 2D Axisymmetric Pavement Structure and Loading Condition


14

0 2 4 6 8 10
-4.80E-01

Maximum Vertical Deflection, mm -4.85E-01

-4.90E-01

-4.95E-01

-5.00E-01

-5.05E-01

-5.10E-01
Model Width, m

(b) Sensitivity Analysis of the Model Radius for the Maximum Vertical Deflection at the
Loading Center

(c) Vertical Deflection History at the Loading Center


Figure 4. The 2D Axisymmetric Pavement FE Model with the Dynamic Load

The 2D axisymmetric model is essentially a cylindrically shaped model. The radial width
represents neither horizontal nor vertical pavement length as in the 3D model. Hence the
boundary effects of the lateral boundary need to be eliminated. A sensitivity analysis was
performed to obtain the required radial width which can guarantee both efficiency and accuracy
15

(Li et al., 2017). In the sensitivity analysis, the pressure magnitude P and the cycle T of the
applied dynamic load are 0.7 MPa psi and 0.03 s. The maximum vertical deflections at the
loading center were checked since it serves as an essential index in this sensitivity analysis.
Figure 4(b) shows the maximum vertical deflections at the loading center in the
cylindrical models of different radii. It illustrates that the model radius indeed affects the model
results. With the increase of the width, the result of the maximum vertical deflection converges,
which indicates the weakening of the boundary effects. The relative error between models with
6.35-m (250-inch) and 7.62-m (300-inch) radii is less than 0.2%. Hence, the 7.62 m is used as
the radius of the 2D axisymmetric models in this study.
Figure 4(c) shows the vertical deflection history at the loading center in the 2D
axisymmetric pavement model under the dynamic load. It reflects the response of the loading
center from load initiation to the end. Similar to the deflection history in the 3D models, the
maximum vertical deflection and the time to reach it can be seen as the characteristics of the
pavement structure.
From Figure 3(a) and Figure 4(c), the deflection history of the points at the pavement
surface under the moving load and dynamic load are of similar shapes, which can be
characterized by the maximum deflection, the time to reach the maximum deflection, etc. The
maximum deflection and the time to reach the maximum deflection reflect the structural and
material properties of the pavement including the layer thickness and the viscoelastic properties
of the layer material: the magnitudes of its modulus and phase angles. These can be used to
determine the deterioration condition as described in previous sections. Therefore, in this study,
the required responses are the maximum deflection and the time to reach the maximum
deflection.
3.3 Latin Hypercube Sampling
The Latin hypercube sampling (LHS) was proposed by Mckay et al. (McKay et al. 1979) to
generate sample points from a multidimensional distribution. In LHS, the number of sample
points is decided and the sample space is partitioned before the sampling to provide random and
uniformly distributed sample points. Compared with other statistical sampling methods, the LHS
has the superiority of simplicity and desirable sampling performance. In this study, the LHS
sampling method was applied to generate 400 two-dimensional sample points of the pressure
magnitude and cycle. These points were used as inputs for dynamic pressures in 2D
16

axisymmetric models. As shown in Figure 5, the sampling ranges of the pressure magnitude and
the cycle of the dynamic load are [0.34, 1.04] (MPa) and [0, 0.5] (s).
1.04

0.9
Pressure Magnitude P, MPa

0.76

0.62

0.48

0.34
0 0.1 0.2 0.3 0.4 0.5
Cycle T, s
Figure 5. The LHS for the Pressure Magnitude and Cycle

3.4 Kriging Model


In the problems of real structures, the relationship between the structural responses and the input
structural parameters is sometimes hard to determine since it may involve complicated and time-
consuming analytical and numerical models. In this study, the equivalent loading magnitude and
cycle are difficult to obtain merely based on the 400 2D axisymmetric FE models. Therefore, the
Kriging model was applied to serve as a surrogate model to provide an explicit relationship
between the structural parameters (pressure magnitude and cycle) and the structural responses
(maximum deflection and time to reach the maximum deflection). The Kriging model is a half-
parameterized interpolation model which includes parametric and non-parametric parts (Qin et
al., 2018a). It provides an explicit relationship for the implicitly-related parameters. Different
from the curve fitting technique or models based on simple functions (i.e. the response surface
model which is based on a polynomial function), the Kriging model considers the global trend as
well as the random errors of data samples (Liu et al., 2014; Qin et al., 2018a). The relationship
17

between the structural responses and input parameters can be expressed as in Equation 4 (Qin et
al., 2018a).
n
Y ( x)    i f i ( x)  Z ( x) (4)
i 1

where Y = structural responses;


x = structural parameters;
 = Kriging model coefficients;
f = simple fitting function (i.e. constant, linear function or quadratic function, etc.);
Z = local deviations in the sample space;
n = the number of structural responses.

As shown in Equation 4, the structural responses Y were obtained from FE models and
the Kriging model coefficients  control the global trend. The training of the Kriging model can
be summarized as solving for the Kriging coefficients in the search for the best linear unbiased
estimate Yˆ , which are the linear combinations of structural responses Y .
In this study, the results of 400 FE models were separated into 300 for Kriging model
training and 100 for Kriging model validation. Figure 6 shows the results of the Kriging model
validation, which is the comparison of structural responses obtained from the FE models and
predicted by the trained Kriging model. The for both responses are close to 1, which reflects high
accuracy of the trained Kriging model.
18

12

Predicted by Kriging Model, mm


10

6
R 2  0.9994
4
Data
2
Line of Equality
0
0 2 4 6 8 10 12
Obtained from FE Models, mm
(a) The Maximum Deflection
0.3
Predicted by Kriging Model, s

0.25

0.2

0.15
R 2  0.9996
0.1
Data
0.05
Line of Equality
0
0 0.05 0.1 0.15 0.2 0.25 0.3
Obtained from FE Models, s

(b) Time to Reach the Maximum Deflection


Figure 6. Comparison of Model Responses Obtained from FE Models and Predicted by
Kriging Model

3.5 Particle Swarm Optimization


19

Once the relationship between the structural responses and input parameters is determined, the
corresponding structural parameters can be obtained for a given structural response. The Particle
Swarm Optimization (PSO) algorithm was applied in searching for the optimal structural
parameters using the obtained Kriging model. PSO is a typical swarm intelligence, which
simulates the swarm behaviors of animals such as fishes, birds and bees. When those animals are
searching for food, the individuals exchange information with each other and the environment
(Lu et al., 2012; Qin et al., 2018b). Similarly, one particle moves in a multi-dimensional space,
decides its moving distance and direction based on the movement of its own and other particles,
and finally finds the optimal location. Compared with other artificial intelligence methods, PSO
has the advantages of simple concept and expression, which is as in Equation 5 and Equation 6
(Shabbir and Omenzetter, 2015).
x i (t  1)  x i (t )  v i (t  1) (5)
in which
v i (t  1)   v i (t )  c1  rand 1  [ pbest i (t )  x i (t )]  c 2  rand 2  [ gbest (t )  x i (t )] (6)
where x i = the position of the i-th particle;
v i = the velocity of the i-th particle;
pbest i = the best position of the i-th particle in its past movements;
gbest = the best position of all particles;
 = the inertial weight;
rand 1 , rand 2 =random numbers in the range [0,1];
t = the generation number;
c1 , c 2 = the cognition and social coefficients.

It can be seen that the position and moving velocity of the particle for each updating
generation are determined by the moving history of its own and the states of all particles. Finally,
all particles approach to one location, which represents the optimal solution of the problem. In
this study, the particle means structural parameters and the best position and can be evaluated
using the objective function expressed in Equation 7. By minimizing the difference between the
structural responses of the 3D FE models and the Kriging model which is trained using the 2D
20

axisymmetric FE model results, the optimal structural parameters are the equivalent structure
parameters.
2
F ( x)  W  w( x) (7)

where x = structural parameters;


W = structural responses of the 3D FE model;
w( x) = structural responses of the Kriging model using the structural parameters x .

3.6 Model Updating


The model updating algorithm in this study can be summarized as Figure 7.

LH Sampling The moving load with


Dynamic load
the specified speed V

Dynamic load with


magnitude P and cycle T Kriging surrogate model 3D FE model

2D Axisymmetric FE Maximum deflection and Maximum deflection


model time to reach it and time to reach it

Maximum deflection and


time to reach it

Objective function value


PSO

No
Stop criterion

Yes

Equivalent dynamic load

Figure 7. Flow Chart of the FE Model Updating


21

Step 1: Build 3D FE models reflecting real pavement structures in ABAQUS and record
the structural responses under moving loads.
Step 2: Build 2D axisymmetric FE models using structural parameters selected from LHS
and record the corresponding structural responses.
Step 3: Train and validate a Kriging model using structural parameters and responses
from Step 2.
Step 4: Find the optimal structural parameters in the Kriging model from Step 3 using the
objective function shown in Equation 7 and the PSO algorithm.
The structural parameters that were obtained including the loading magnitude and cycle
time are presented in Table 3. These parameters are the equivalent loading magnitudes and
cycles of the dynamic loads for the moving load at different speeds.

Table 3. Equivalent Loading Magnitudes and Cycles of Dynamic Loads for Moving Loads
Equivalent Dynamic Load Parameters
Speed of the Moving Load, km/h Pressure Magnitude P,
Cycle T, s
MPa
40.23 0.4886 0.3163
56.33 0.4841 0.2219
72.42 0.4748 0.1687
80.47 0.4750 0.1500
96.56 0.4821 0.1204
112.65 0.4825 0.1018
128.75 0.4800 0.0856
160.93 0.4697 0.0635

In order to ensure the accuracy of the model updating, these equivalent parameters were
used in the newly-built 2D axisymmetric FE models to compare the structural responses with
those in the 3D FE models. Results are shown in Figure 8. The high values of validates the
success of the model updating.
22

0.5

Results of 2D Axisymmetric Models, mm


0.48

0.46

0.44

0.42
R 2  0.9734
0.4 Data
Line of Equality
0.38
0.38 0.4 0.42 0.44 0.46 0.48 0.5
Results of 3D FE Models, mm
(a) The Maximum Deflection

0.2
Results of 2D Axisymmetric Models, s

0.15

0.1

0.05 R 2  0.9998

Data
Line of Equality
0
0 0.05 0.1 0.15 0.2
Results of 3D FE Models, s
23

(b) Time to Reach the Maximum Deflection


Figure 8. Comparison of Model Responses of 3D FE Models and 2D Axisymmetric FE
Model

4 Evaluation of Pavement Deterioration Condition Using Surface Deflection


Since the equivalent dynamic load magnitude and cycle time have been determined for the
moving loads at different speeds, the 2D axisymmetric FE models can be applied to study the
corresponding structural responses in different deterioration conditions and under loads at
different speeds.
For different deterioration conditions, the modulus and phase angle of the pavement
surface material were adjusted to follow the trends before the formation of macro-cracks as in
the fatigue tests. The values of dynamic modulus and phase angle were assumed to be measured
at the frequency of 10 Hz, which is one of the testing frequencies in the dynamic modulus test
and the fatigue test for asphalt mixtures (Bonaquist 2008; Poulikakos et al., 2015). The values
and ratios of the dynamic modulus and the phase angle for pavement surface materials at
different deterioration conditions are presented in Table 4. With the deteriorating condition of the
material, the dynamic modulus decreases while the phase angle increases.

Table 4. Dynamic Modulus and Phase Angle of Pavement Surface Material at Different
Deterioration Levels

Deterioration Level i E* / E*
i 0
E * , MPa
i
i /  0 i , 
0 1.0 2.99E+3 1.0 29.44
1 0.9 2.68E+3 1.1 32.38
2 0.8 2.39E+3 1.3 38.27
3 0.7 2.09E +3 1.4 41.21
4 0.6 1.79E +3 1.6 47.10
5 0.5 1.49E +3 1.8 52.99

To implement the material properties in Table 4 into ABAQUS, the Prony Series
coefficients were obtained by minimizing the differences between the dynamic modulus E * and

phase angle  listed in Table 4 and predicted by Equation 8 and Equation 9 (Zhang et al., 2016).
24

E * ( )  [ E ( )] 2  [ E ( )] 2 (8)

 E ( ) 
  tan 1   (9)
 E ( ) 
in which
N
 2 i2 E i
E ( )  E    (10)
i 1 1    i
2 2

N
 i E i
E ( )   (11)
i 1 1    i
2 2

where E ( ) = the storage modulus;


E ( ) = the loss modulus;
 = the loading frequency;
E  = long-term modulus;
E i = Prony regression coefficients;
 i = relaxation time;
N = number of Prony series terms.

The pavement structures with surface materials representing 6 deterioration conditions


were applied with 8 dynamic loads representing the moving loads at 8 different speeds. 48 2D
axisymmetric FE models in total were built and the deflection histories at the loading center were
analyzed. Following the idea of the lag angle (Deng et al., 2019) which is defined from the
dimension of the deflection basins of the pavement surface under moving loads in 3D FE
models, a similar concept was proposed to evaluate the deterioration condition of the pavement
using its deflected profile in the 2D axisymmetric model.
25

Figure 9. Normalized Load and Deflection Histories under the Moving Load of 50 mph

Figure 9 shows the histories of normalized load and deflection under the dynamic load of
equivalent magnitude and cycle time for a moving load of 80.47 km/h. There is an obvious time
lag between the maximum load and deflection in the plot. In order to utilize this characteristic to
evaluate the deterioration condition of the flexible pavement, a similar term was defined as in
Equation 12 and is called the lag angle since it is also from the deflection of the pavement
surface.
t
  2 (12)
T
where  = lag angle of the pavement (radius);
t = the time lag between the maximum load and deflection;
T = cycle time of the deflection basin.

Table 5 presents the calculated lag angles for a total of 48 cases and Figure 10 shows the
correlations between the lag angle of the pavement and the phase angle of the surface material
under different moving loads. That figure shows that the lag angle increases with the increase of
the phase angle and the speed of the moving load. Such trend is similar to and can also be found
in the lag angle calculated from deflection basins (Deng et al., 2019). It indicates that a causal
26

relation between the material properties of the flexible pavement surface and the structural
response of the pavement.

Table 5. Lag Angle (  ) of Pavements of Different Deterioration Conditions under Moving


Loads of Different Speeds
Phase Speed of the Moving Load, km/h
Angle,  40.23 56.33 72.42 80.47 96.56 112.65 128.75 160.93
29.44 19.00 24.27 26.44 27.39 29.81 32.99 37.38 43.35
32.38 19.01 24.26 26.48 27.33 29.94 33.21 37.58 43.87
38.27 19.83 25.50 27.86 28.83 31.44 34.67 38.98 45.18
41.21 19.86 25.62 28.14 29.13 31.64 34.90 39.30 45.44
47.10 21.41 27.48 30.04 31.08 33.62 36.82 41.06 47.22
52.99 23.14 29.15 31.64 32.57 34.91 38.06 42.17 48.13

Figure 10. Relationship between Phase Angle, Lag Angle and Moving Speed

5 Sensitivity Analysis of Lag Angle


As a parameter determining the pavement deterioration condition, the lag angle should be
sensitive to pavement properties and show explicit relations with them. The relations between the
lag angle and pavement properties such as the layer thickness and modulus were checked by
adjusting pavement properties in the FE model and calculating the lag angle of the pavement.
27

Figure 11 shows the relations between the lag angle of the pavement and four pavement
properties – the surface layer thickness, the base layer thickness, the base layer modulus and the
subgrade modulus. Values of properties were taken within normal ranges (NCHRP, 2004). It can
be seen that the lag angle increases as the subgrade modulus or the surface layer thickness
increases. Opposite trends can be observed in the cases in which the modulus or the thickness of
the base layer was adjusted.

29

28.5
Lag Angle of Pavement, degree

28

27.5

27

26.5

26

25.5

25
120 160 200 240 280 320 360
Base Modulus, MPa
(a) Base Layer Modulus
28

31
30.5
Lag Angle of Pavement, degree 30
29.5
29
28.5
28
27.5
27
26.5
26
25.5
0 50 100 150 200 250
Subgrade Modulus, MPa
(b) Subgrade Modulus

34

32
Lag Angle of Pavement, degree

30

28

26

24

22

20

18
40 60 80 100 120 140 160
Surface Thickness, mm

(c) Surface Layer Thickness


29

28

Lag Angle of Pavement, degree

27.5

27

26.5
80 130 180 230
Base Thickness, mm

(d) Base Layer Thickness


Figure 11. Relationships between Lag Angle and Pavement Properties

An explicit and monotonic relation indicates a causal correlation between the lag angle
and the pavement property. Also, it shows a potential use of the lag angle to determine the
pavement deterioration condition and other pavement properties in an inverse analysis.

6 Conclusions and Future Work


This paper proposes a method of evaluating the microcracking deterioration condition of flexible
pavements using their deflection profiles under moving loads. Also, an AI-involved FE model
updating algorithm is proposed to construct the equivalent 2D axisymmetric FE model under a
dynamic load to represent the 3D FE model under a moving load. The major contributions are
summarized as follows.
 The 3D FE analysis shows that the moving loads with higher speeds result in smaller
deflections and shorter response durations.
30

 The equivalent load magnitude and cycle time of the dynamic load for the moving load
show that the moving loads with higher speeds have equivalent dynamic loads of smaller
cycle times. Such trend is not obvious in the equivalent load magnitude.
 The 2D axisymmetric FE analysis shows that there is a time lag between the load and
deflection peaks. The time lag is used to define a new term “lag angle” to reflect the
structural response of the flexible pavement under the moving load.
 The deterioration degree of the flexible pavement affects the value of the lag angle. With
the increase of the deterioration severity, the lag angle increases. The increase of the phase
angle of the material is attributed to the initiation and propagation of micro-cracks in the
material.
 The speed of the moving load affects the value of the lag angle. With the increase of the
moving speed, the lag angle increases.
 The lag angle shows explicit and monotonic relations with pavement properties. With the
increase of the subgrade modulus or the surface layer thickness, the lag angle increases.
With the increase of the base modulus or the base layer thickness, the lag angle decreases.

The relationship between the lag angle, the phase angle and the speed of the moving load
is built based on the limited 48 cases. In the future, more structural and material parameters will
be included for a comprehensive analysis of the lag angle. In addition, it is planned to compare
the lag angle defined in this study with other measurements made in the field and indices in the
literature for the application of the lag angle in evaluating the deterioration conditions in flexible
pavements.
31

Reference
[1] ABAQUS. (2012). Abaqus 6.12 Documentation and User Manual. Dassault Systèmes
Simulia Corp.
[2] Al-Qadi, I. L., Loulizi, A., Janajreh, I., and Freeman, T. E. (2002). Pavement response to
dual tires and new wide-base tires at same tire pressure. Transportation Research Record,
1806(1), 38-47.
[3] Ahmed, T. M., and Khalid, H. A. (2015). A new approach in fatigue testing and evaluation
of hot mix asphalt using a dynamic shear rheometer. In Proc., 6th Int. Conf. on Bituminous
Mixtures and Pavements, pp. 351-359.
[4] Al-Qadi, I. L., Xie, W., and Elseifi, M. A. (2008). Frequency determination from vehicular
loading time pulse to predict appropriate complex modulus in MEPDG. Asphalt Paving
Technology-Proceedings, 77, 739.
[5] Bonaquist, R. F. (2008). Refining the simple performance tester for use in routine practice.
Transportation Research Board, National Research Council, Washington, D.C.
[6] Boudabbous, M., Millien, A., Petit, C., and Neji, J. (2013). Energy approach for the fatigue
of thermoviscoelastic materials: application to asphalt materials in pavement surface layers.
International journal of fatigue, 47, 308-318.
[7] Cocurullo, A., Airey, G. D., Collop, A. C., and Sangiorgi, C. (2008). Indirect tensile versus
two-point bending fatigue testing. In Proceedings of the Institution of Civil Engineers-
Transport, Vol. 161, No. 4, pp. 207-220, Thomas Telford Ltd.
[8] Carpenter, S. H., Ghuzlan, K. A., and Shen, S. (2003). Fatigue endurance limit for highway
and airport pavements. Transportation research record, 1832(1), 131-138.
[9] Chai, G., S. Manoharan, A. Golding, G. Kelly, and S. Chowdhury. (2016). Evaluation of the
Traffic Speed Deflectometer Data Using Simplified Deflection Model. Transportation
Research Procedia, 14:3031-3039.
[10]Diefenderfer, B. K. (2010). Investigation of the Rolling Wheel Deflectometer as a Network-
level Pavement Structural Evaluation Tool. No. VTRC 10-R5, FHWA/VTRC 10-R5. Virginia
Transportation Research Council.
[11]Di Benedetto, H., de La Roche, C., Baaj, H., Pronk, A., and Lundstrom, R. (2003). Fatigue
of bituminous mixtures: different approaches and RILEM group contribution. In Sixth
International RILEM Symposium on Performance Testing and Evaluation of Bituminous
32

Materials, pp. 15-38. RILEM Publications SARL.


[12]Deng, Y., Luo, X., Gu, F., Zhang, Y., and Lytton, R. L. (2019). 3D simulation of deflection
basin of pavements under high-speed moving loads. Construction and Building Materials,
226, 868-878.
[13]Di Benedetto, H., Nguyen, Q. T., and Sauzéat, C. (2011). Nonlinearity, heating, fatigue and
thixotropy during cyclic loading of asphalt mixtures. Road Materials and Pavement Design,
12(1), 129-158.
[14]Elbagalati, O., Elseifi, M. A., Gaspard, K., and Zhang, Z. (2016). Prediction of In-service
Pavement Structural Capacity Based on Traffic-speed Deflection Measurements. Journal of
Transportation Engineering, 142(11), 04016058.
[15]Elbagalati, O., Elseifi, M. A., Gaspard, K., and Zhang, Z. (2017). Implementation of the
Structural Condition Index into the Louisiana Pavement Management System Based on
Rolling Wheel Deflectometer Testing. Transportation Research Record, 2641(1): 39-47.
[16]Golding, A., Chowdhury, S., Chai, G., and Manoharan, S. (2017). A Study of the Structural
Performance of Flexible Pavements Using Traffic Speed Deflectometer. Journal of Testing
and Evaluation, 46(3): 1280-1289.
[17]Gu, F., Luo, X., Luo, R., Lytton, R. L., Hajj, E. Y., and Siddharthan, R. V. (2016).
Numerical modeling of geogrid-reinforced flexible pavement and corresponding validation
using large-scale tank test. Construction and Building Materials, 122, 214-230.
[18]Gu, F., Luo, X., Zhang, Y., Chen, Y., Luo, R., and Lytton, R. L. (2018). Prediction of
geogrid-reinforced flexible pavement performance using artificial neural network approach.
Road Materials and Pavement Design, 19(5), 1147-1163.
[19]Garcia, G., and Thompson, M. R. (2008). Strain and pulse duration considerations for
extended-life hot-mix asphalt pavement design. Transportation research record, 2087(1), 3-
11.
[20]Huang, Y. H. (1993). Pavement Analysis and Design. Prentice Hall, Englewood Cliffs, New
Jersey.
[21]Kim, Y. R., Little, D. N., and Lytton, R. L. (2003). Fatigue and healing characterization of
asphalt mixtures. Journal of Materials in Civil Engineering, 15(1), 75-83.
[22]Krishnamoorthy, C. S., and Rajeev, S. (2018). Artificial intelligence and expert systems for
engineers. CRC press, Boca Raton, FL.
33

[23]Luo, X., Birgisson, B. and Lytton, R. L. (2020). Kinetics of healing of asphalt mixtures.
Journal of Cleaner Production, Vol. 252, p.119790.
[24]Lu, P., Chen, S., and Zheng, Y. (2012). Artificial intelligence in civil engineering.
Mathematical Problems in Engineering, 2012.
[25]Ling, M., Luo, X., Hu, S., Gu, F., and Lytton, R. L. (2017). Numerical modeling and
artificial neural network for predicting J-integral of top-down cracking in asphalt pavement.
Transportation Research Record, 2631(1), 83-95.
[26]Liu, Y., Wang, D. J., Ma, J., and Li, Y. (2014). Performance comparison of two meta-model
for the application to finite element model updating of structures. J. Harbin Inst. Tech, 21,
68-78.
[27]Li, M., Wang, H., Xu, G., and Xie, P. (2017). Finite element modeling and parametric
analysis of viscoelastic and nonlinear pavement responses under dynamic FWD loading.
Construction and Building Materials, 141, 23-35.
[28]McKay, M. D., Beckman, R. J., and Conover, W. J. (1979). Comparison of three methods
for selecting values of input variables in the analysis of output from a computer code.
Technometrics, 21(2), 239-245.
[29]Moreno-Navarro, F., and Rubio-Gámez, M. C. (2016). A review of fatigue damage in
bituminous mixtures: Understanding the phenomenon from a new perspective. Construction
and Building Materials, 113, 927-938.
[30]National Cooperative Highway Research Program (NCHRP). (2004). Guide for
Mechanistic-empirical Design of New and Rehabilitated Structures. Final Rep. for Project
1-37A, Transportation Research Board, National Research Council, Washington, D.C.
[31]Nasimifar, M., Thyagarajan, S., Siddharthan, R. V., and Sivaneswaran, N. (2016). Robust
deflection indices from traffic-speed deflectometer measurements to predict critical
pavement responses for network-level pavement management system application. Journal of
Transportation Engineering, 142(3), 04016004.
[32]Nasimifar, M., Thyagarajan, S., and Sivaneswaran, N. (2017). Backcalculation of flexible
pavement layer moduli from traffic speed deflectometer data. Transportation Research
Record, 2641(1), 66-74.
[33]Nasimifar, M., Thyagarajan, S., and Sivaneswaran, N. (2018). Computation of Pavement
Vertical Surface Deflections from Traffic Speed Deflectometer Data: Evaluation of Current
34

Methods. Journal of Transportation Engineering, Part B: Pavements, 144(1), 04018001.


[34]Pronk, A. C., and Hopman, P. C. (1991). Energy dissipation: the leading factor of fatigue. In
Highway research: Sharing the benefits, pp. 255-267. Thomas Telford Publishing.
[35]Poulikakos, L. D., Pittet, M., Dumont, A. G., and Partl, M. N. (2015). Comparison of the
two-point bending and four-point bending test methods for aged asphalt concrete field
samples. Materials and Structures, 48(9), 2901-2913.
[36]Qabur, A. (2018). Fatigue Characterization of Asphalt Mixes with Polymer Modified
Asphalt Cement, Master thesis, University of Waterloo, Waterloo, Ontario, Canada.
[37]Qin, S., Zhou, Y. L., Cao, H., and Wahab, M. A. (2018a). Model updating in complex bridge
structures using kriging model ensemble with genetic algorithm. KSCE Journal of Civil
Engineering, 22(9), 3567-3578.
[38]Qin, S., Zhang, Y., Zhou, Y. L., and Kang, J. (2018b). Dynamic model updating for bridge
structures using the kriging model and PSO algorithm ensemble with higher vibration
modes. Sensors, 18(6), 1879.
[39]Reese, R. (1997). Properties of aged asphalt binder related to asphalt concrete fatigue life.
Journal of the Association of Asphalt Paving Technologists, 66.
[40]Schapery, R. A. (1961). A Simple Collocation Method for Rutting Viscoelastic Models to
Experimental Data. Report GALCIT SM 61-23A, California Institute of Technology,
Pasadena, CA.
[41]Shen, S., Airey, G. D., Carpenter, S. H., and Huang, H. (2006). A dissipated energy
approach to fatigue evaluation. Road materials and pavement design, 7(1), 47-69.
[42]Saha, S., Gu, F., Luo, X., and Lytton, R. L. (2018). Use of an artificial neural network
approach for the prediction of resilient modulus for unbound granular material.
Transportation Research Record, 2672(52), 23-33.
[43]Shabbir, F., and Omenzetter, P. (2015). Particle swarm optimization with sequential niche
technique for dynamic finite element model updating. Computer‐Aided Civil and
Infrastructure Engineering, 30(5), 359-375.
[44]Tapsoba, N., Sauzéat, C., and Benedetto, H. D. (2012). Analysis of fatigue test for
bituminous mixtures. Journal of Materials in Civil Engineering, 25(6), 701-710.
35

[45]Tapsoba, N., Sauzéat, C., Di Benedetto, H., Baaj, H., and Ech, M. (2015). Three-
dimensional analysis of fatigue tests on bituminous mixtures. Fatigue & Fracture of
Engineering Materials & Structures, 38(6), 730-741.
[46]Ulloa, A., Hajj, E. Y., Siddharthan, R. V., and Sebaaly, P. E. (2013). Equivalent loading
frequencies for dynamic analysis of asphalt pavements. Journal of Materials in Civil
Engineering, 25(9), 1162-1170.
[47]Wang, L., Hou, Y., Zhang, L., and Liu, G. (2017). A combined static-and-dynamics
mechanics analysis on the bridge deck pavement. Journal of Cleaner Production, 166, 209-
220.
[48]Wang, H., and Li, M. (2016). Comparative study of asphalt pavement responses under FWD
and moving vehicular loading. Journal of Transportation Engineering, 142(12), 04016069.
[49]Zhang, Y., Birgisson, B., and Lytton, R. L. (2016). Weak form equation–based finite-
element modeling of viscoelastic asphalt mixtures. Journal of Materials in Civil
Engineering, 28(2), 04015115.
36

AUTHORSHIP STATEMENT
Manuscript Title:

Evaluation of Flexible Pavement Deterioration Conditions Using Deflection Profiles Under


Moving Loads

All persons who meet authorship criteria are listed as authors, and all authors certify that they
have participated sufficiently in the work to take public responsibility for the content, including
participation in the concept, design, analysis, writing, or revision of the manuscript. Furthermore,
each author certifies that this material or similar material has not been and will not be submitted
to or published in any other publication before its appearance in the Transportation Geotechnics.

Authorship Contributions:

Conception and design of study: Y. Deng, X. Luo, and R. L. Lytton


Finite element simulation and model updating: Y. Deng and Y. Zhang
Analysis and/or interpretation of data: Y. Deng, X. Luo, Y. Zhang, and R. L. Lytton
Drafting the manuscript: Y. Deng
Revising the manuscript critically for important intellectual content: Y. Deng, X. Luo, and R. L.
Lytton

Acknowledgements

All persons who have made substantial contributions to the work reported in the manuscript
(e.g., technical help, writing and editing assistance, general support), but who do not meet the
criteria for authorship, are named in the Acknowledgements and have given us their written
permission to be named. If we have not included an Acknowledgements, then that indicates that
we have not received substantial contributions from non-authors.
37

Declaration of Interest

The authors declare that they have no conflict of interest.

You might also like