Sie sind auf Seite 1von 10

Construction and Building Materials 64 (2014) 360–369

Contents lists available at ScienceDirect

Construction and Building Materials


journal homepage: www.elsevier.com/locate/conbuildmat

Evaluation of Witczak E predictive models for the implementation


of AASHTOWare-Pavement ME Design in the Kingdom of Saudi Arabia
Ahmed M. Khattab a, Sherif M. El-Badawy a,⇑, Al Abbas Al Hazmi b, Mahmoud Elmwafi c
a
Mansoura University, Public Works Engineering Department, Faculty of Engineering, Mansoura University, Mansoura 35516, Egypt
b
General Directorate for Material and Research, Riyadh 11178, Saudi Arabia
c
Mansoura University, Public Works Engineering Department, Mansoura 35516, Egypt

h i g h l i g h t s


 NCHRP1-37A yielded better E values compared to NCHRP 1-40D model at all binder characterization input levels.

 NCHRP1-37A E model with Level 3 binder characterization produced the most accurate and the least biased estimates.

 Because of the hot desert climate in KSA, care should be taken when using such models for E predictions.

a r t i c l e i n f o a b s t r a c t

Article history: AASHTOWare-Pavement ME Design is the production version of the Mechanistic-Empirical Pavement
Received 23 January 2014 Design Guide (MEPDG) which includes two different models for hot mix asphalt (HMA) dynamic modulus
Received in revised form 27 March 2014 (E) prediction. The first model is the NCHRP 1-37A while the second one is the NCHRP 1-40D model. This
Accepted 4 April 2014
paper focuses on the implementation of the ME Design in the Kingdom of Saudi Arabia by developing a
Available online 4 May 2014
database containing mechanical properties of a wide variety of HMA mixtures. Then checking the
predictive power of both models using the developed database. Laboratory E tests were conducted on
Keywords:
25 different HMA mixtures. Dynamic Shear Rheometer and Brookfield tests were also conducted on
MEPDG
AASHTOWare
the binders. Results showed that the performance of the investigated models varied by temperature
Dynamic Modulus and the binder characterization method. The results also showed that the NCHRP 1-37A E model along
Rutting with Level 3 binder inputs yielded the most accurate and least biased E estimates.
Temperature Ó 2014 Elsevier Ltd. All rights reserved.
Modified binder
Witczak
DSR
Brookfield

1. Introduction which is the production version of the Mechanistic Empirical


Pavement Design Guide (MEPDG) represents a major change in
For decades, pavement engineers have tried to find methods for the philosophy of pavement design. It computes stresses, strains,
measuring the performance of asphalt concrete pavements by and deflections within a pavement system and then predicts,
predicting distresses developed in pavement systems due to traffic through empirical, however, accurate models various distresses
and environmental conditions throughout the pavement design in pavements including rutting, cracking, and, roughness during
life. The NCHRP 9-19 project studied several mix parameters and the pavement service life [2].
concluded that the dynamic modulus (E) of the asphalt mixture Ministry of Transport (MOT) in the Kingdom of Saudi Arabia
is strongly correlated to pavement performance over a diverse (KSA) is gearing towards the implementation of the ME-Design.
range of traffic and climatic conditions [1]. AASHTOWare This is expected to improve the flexible pavement design process.
Pavement-ME Design (from here on will be called ‘‘ME-Design’’) It will also provide a better capability for the prediction of
pavement lifetime maintenance needs. However, using ME-Design
for the design of asphalt pavements requires, among others, the
⇑ Corresponding author. Tel.: +20 100 018 3519.
characterization of the hot mix asphalt (HMA) materials as well
E-mail addresses: eng.ahmed.khattab.1989@gmail.com (A.M. Khattab),
sbadawy@mans.edu.eg (S.M. El-Badawy), abas.tiger@hotmail.com (A.A. Al Hazmi),
as the unbound layers and subgrade soil. The ME-Design provides
mmewafi1@excite.com (M. Elmwafi). three hierarchical levels of inputs (i.e. Level 1, Level 2, and Level 3).

http://dx.doi.org/10.1016/j.conbuildmat.2014.04.066
0950-0618/Ó 2014 Elsevier Ltd. All rights reserved.
A.M. Khattab et al. / Construction and Building Materials 64 (2014) 360–369 361

The selection of a particular level of input depends on the amount for the investigated mixes taking into account the impact of the
of information available to the designer and the criticality of the binder characterization input level as well as the temperature.
project. For all three input levels, E is the main parameter to com-
pute the stress-strain characteristics of the asphalt layers at the
critical locations due to traffic loads taking into account the impact 3. Investigated mixtures
of truck speeds and environmental conditions. Dynamic modulus
values for Level 1 are measured in the laboratory at selected com- KSA has been using the Superpave mix design since 2005 [21].
binations of temperatures and loading frequencies. However, it is The current KSA Superpave coding system is shown in Table 1. It
not always feasible to conduct this test because of the availability is based on the region, anticipated project traffic level expected
of the required equipment and time constraints. For Levels 2 and 3 on the design lane over a 20-year period, aggregate consensus
HMA stiffness characterization, two different E predictive models properties, and binder performance grade (PG). For example a
were implemented in the software for E prediction at the antici- Superpave code 12R8C4Y means 12.5 mm Nominal Maximum
pated temperature and loading rate. Both models were developed Aggregate Size (NMAS) mix, project is located in Hail Region,
by Witczak and his associates based on a large E database. The design Equivalent Single Axle Load (ESAL) less than 30 million, this
first model is called NCHRP 1-37A Viscosity (g)-based model while mix will be located at a depth less than 100 mm from the surface,
the other is the NCHRP 1-40D binder shear modulus (Gb)-based and the mix binder is PG 76-10.
model [3–9]. Both models predict dynamic modulus of HMA at To cover a wide range of mixes in the kingdom, 25 Superpave
the anticipated temperature and frequency of loading as a function mixes were collected from different ongoing construction projects.
of mix volumetric properties, mix aggregate gradation, and a bin- The selected mixes cover the central, northern, southern and,
der stiffness term. Traditional binder viscosity is the parameter eastern regions. Out of the 25 mixes, 14 were mixed in the labora-
used for the binder stiffness in the NCHRP 1-37A model whereas tory while the remaining 11 mixes were plant-produced mixes.
binder shear modulus and phase angle (d) are the parameters used The laboratory mixes are characterized by (C), (M), and (F) which
for the binder characterization in the NCHRP 1-40D model [3–5,7– mean coarse gradation mix, medium gradation mix, and fine gra-
12]. Similarly, Levels 1 and 2 inputs for the binder characterization dation mix, respectively. Fig. 1 shows the location of the projects
require laboratory measurements based on the binder grading from which these mixtures were collected as well as the different
system. For Superpave performance grade binders, laboratory- climatic regions in KSA. The investigated mixes contain virgin and
measured Gb and d values at different temperatures and one angu- modified binders as shown in Table 2. The Job Mix Formula volu-
lar loading frequency of 10 rad/s (1.59 Hz) are the required inputs. metric properties and gradation of the investigated mixtures are
For conventional binder grades, traditional asphalt testing results, summarized in Table 3.
such as penetration, ring and ball softening point, absolute and
kinematic viscosities, and Brookfield viscosities at different tem-
3.1. Dynamic modulus measurement (Level 1)
peratures, are the required inputs. For Level 3 binder inputs, users
are only required to select the binder grade based on one of the
Each E specimen was compacted using Controls Gyratory Com-
binder grading systems (i.e., Superpave performance grade, con-
pactor (CGC) model ICT 250 to achieve cylindrical specimens of
ventional penetration grade, or conventional viscosity grade) [2].
150 mm in diameter and 170 mm in height. The samples were
Recent studies in Idaho, Argentina, Louisiana, Florida, North
trimmed from the top and bottom to reach a final height of
Carolina, and Arizona [3–5,11,13,14,15,16] reported that the
150 mm. Each specimen was then cored in the middle to produce
NCHRP 1-37A model produced better E predictions compared to
a final diameter of 100 mm. Most of the dynamic modulus tests
the NCHRP 1-40D model, especially at the high temperatures. Sev-
were run on at least two replicates per mix. The E tests were con-
eral other studies showed that the NCHRP 1-40D model generally
ducted according to AASHTO T 342-11 at 10, 4.4, 21.1, and
yielded highly biased predictions at high and/or low temperatures
54.4 °C. At each temperature, the tests were conducted at loading
[17–20].
frequencies of 25, 10, 5, 1, 0.5, and 0.1 Hz [22]. Hence, twenty-four
dynamic modulus measurements were determined for each
2. Objectives sample (4 temperatures and 6 frequencies). Asphalt Mixture Per-
formance Tester (AMPT), which meets the AASHTO T 342-11
This research aims at developing a comprehensive E database requirements was used to run all E tests. All tests were conducted
for HMA widely used in KSA. It also aims at evaluating the at the MOT Laboratory, General Directorate for Material and
predictive power of the E models implemented in the ME-Design Research, Quality Control Department in Riyadh, KSA.

Table 1
Coding system of Superpave mixes in the Kingdom of Saudi Arabia.

Mix Geographical Location Traffic Consensus Properties Binder Performance Grade


NMASa Code Region Code Design ESAL Code Depth from the surface Code PG grade Code
9.5 mm 9 Riyadh R1 <0.3 million 1 PG 64-10 A
Makkah R2
Madinah R3
Qasim R4
12.5 mm 12 Eastern R5 <3.0 million 2 <100 mm Y PG 70-10 B
Asir R6
Tabouk R7
19.0 mm 19 Hail R8 <10 million 3 >100 mm N PG 76-10 C
Northern Border R9
25.0 mm 25 Jizan R10 <30 million 4 PG 82-10 D
Najran R11
37.5 mm 37 Baha R12 >30 million 5 Other E
Al-Jouf R13
a
NMAS = Nominal maximum aggregate size.
362 A.M. Khattab et al. / Construction and Building Materials 64 (2014) 360–369

Fig. 1. Location of the selected projects for the investigated mixtures.

Table 2
Investigated mixtures.

MIX ID Aggregates sources (project) MOT code PG Polymer type Polymer percent
grade %
Laboratory mixes
T6F,T6M,T6C Doublization of Taif-Diraab Road 12R1A4Y 64-10 – –
T7F,T7M,T7C Doublization of Taif-Diraab Road 12R1A4Y 76-10 Polybilt 101 5
H6F,H6M,H6C Completion of Dyraab Road to Riyadh-Taif Road 12R8A4Y 64-10 – –
H7F,H7M,H7C Completion of Dyraab Road to Riyadh-Taif Road 12R8A4Y 76-10 Polybilt 101 5
A Construction of Hail – Al Gouf Road (Al Adwaa-Baqaa) 25R8A4Y 64-10 – –
D Construction of Hail – Al Gouf Road (Al Adwaa-Baqaa) 25R8C4Y 76-10 CRMa 8
Field mixes
NH Dhahran Dammam Jubial – Dhahran AbuHadriyah Khafji Hafer Al Baten Rafha 12R5C5Y 76-10 Polybilt 5
Expressway
FW Hail-Madinah Doubled Road 25R8B4Y 70-10 Anglomaks + PPA 1.5 + 0.05
SJ Preventive Road Maintenance Jeddah 25R2C4Y 76-10 EE-2 4
SP Conversion of Jeddah Jizan Road to Expressway S3 25R2C4Y 76-10 CRM 8
HG Hail/Rafha Dual Road (First Stage-Single) 25R8B4Y 70-10 CRMe (RPP) 4
FQ Airport Road Badaya (Jadat Al Jamamil) 25R4C4Y 76-10 EE2 4
HH Reform of the current path of Sabya Abu Arish Road 25R10B4N 70-10 CRM 4
AYUN Roads Leading to the Housing Sites of King Abdullah Bin Abdul Aziz Development 25R10C4Y 76-10 LLDPE + PPAb 2 + 0.5
AJM K Jeddah – Jizan Road Group 1 (Right Side) 25R10C4Y 76-10 EE-2 + PPAc 3.2 + 0.5
AJM D Jeddah – Jizan Road Group 1(Left Side) 25R10C4Y 76-10 EE-2 + PPAd 3.2 + 0.5
BIN JAR Al-Shaqiq Al-Tuwal in Jizan Section 1 25R10B4N 70-10 EE2 + PPA 2 + 0.4
a
CRM = Crumb Rubber.
b
LLDPE = Low Linear Density Polyethylene.
c
EE-2 = Medium Density, Low Molecular-Weight Oxidized Polyethylene with Exceptional Hardness.
d
PPA = Polyphthalamide.
e
RPP = Reversed Phase Polymer.

3.2. Dynamic modulus prediction in ME design (Levels 2 and 3) 3.2.1. NCHRP 1-37A g-based E predictive model
This model predicts mixture stiffness over a wide range of
E values were predicted using both models incorporated in the temperatures and loading frequencies from data that is readily
ME-Design using different binder inputs as per the software available from the volumetric properties of mixture. It was devel-
requirements. This is explained in the following subsections. oped based on a large database containing 2750 measurements
368 A.M. Khattab et al. / Construction and Building Materials 64 (2014) 360–369

Table 6
Evaluation of the predictive procedures in logarithmic space.

ME-Design E models Binder level R2 Se/Sy


NCHRP 1-37A (g-based) Level 1a 0.79 0.46
Level 1b 0.85 0.39
Level 3 0.88 0.35
NCHRP 1-40D (Gb-based) Level 1a 0.70 0.55
Level 1b 0.82 0.43
Level 3 0.84 0.40

sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn 2
i¼1 ei
Se ¼ ð13Þ Fig. 5. Bias in E Prediction Segregated by Temperature Using the NCHRP 1-37A
np
Model.

 2
n  p Se
R2 ¼ 1  ð14Þ
n  1 Sy

where Sy is the standard deviation of the measured E values about


the mean measured; e the error between the predicted and mea-
sured E values; Se the standard error (i.e., standard deviation of
error); R2 is the coefficient of determination.
Both Se/Sy and R2 are measures of the model accuracy (degree of
scatter with reference to the line of equality). The R2 is simply the
square of the correlation coefficient between the measured and
predicted E (higher R2 indicates higher accuracy). The Se/Sy is an
indicator for the relative improvement in accuracy and that means
smaller value points out better accuracy. The line of equality is a
line with slope constrained to unity and intercept constrained to Fig. 6. Bias in E prediction segregated by temperature using the NCHRP 1-40D
zero. Thus, the overall bias in prediction can also be assessed by model.
measuring how close are the slope and intercept of the uncon-
strained linear regression between measured and predicted E val-
ues to 1 and 0, respectively [25]. Fig. 3 and Table 6 show that the R2
for the NCHARP 1-37A ranged from 0.79 to 0.88 with Se/Sy in the These figures indicate that both models yield biased predictions
range of 0.46–0.35. For the NCHRP 1-40D model, the R2 vales were at all temperature ranges. However, the bias was generally more
slightly lower and they ranged from 0.70 to 0.84 with Se/Sy in the pronounced at the very low/low temperatures. At the very high
range of 0.55–0.40. Fig. 4 compares the overall accuracy and bias temperatures, the bias of both models predictions was significantly
of both models. The data in this figure show that, in both models lower at all binder input cases. In general, the degree of bias at the
at the different binder characterization cases produce biased E different temperatures was lower for the NCHRP 1-37A model
estimates. In general, the degree of bias from both models are compared to the NCHRP 1-40D model. Figs. 5 and 6 show that
not significantly different. Results shown in Figs. 3 and 4 along the degree of bias for the NCHRP 1-40D models is much higher
with the goodness-of-fit statistics shown in Table 6 reveal that at the very low temperature compared to the very high tempera-
the NCHRP1-37A model with Level 3 binder characterization gives ture. In addition, the highest degree of bias for this model is
the most accurate (R2 = 0.88, Se/Sy = 0.35) E estimates. associated with Level 1a binder characterization.

6. Effect of temperature on |E| predictions 7. Summary and conclusions

The overall accuracy and bias for the NCHRP 1-37A and NCHRP For implementing ME-Design in the kingdom dynamic modulus
1-40D models segregated by temperature are shown in Figs. 5 tests were conducted on typical HMA mixtures from ongoing
and 6, respectively. paving projects distributed all over the Kingdom. A database

Fig. 4. Overall performances of the investigated models.

Das könnte Ihnen auch gefallen