Sie sind auf Seite 1von 15

TRB Paper 07-3356

Paper Submitted to the Transportation Research Board for Presentation and Publication
at the 86th Annual meeting to be held on January 21-25, 2007 in Washington, D.C..

Condition Assessment of Composite Pavement Systems Using Neural


Network-Based Rapid Backcalculation Algorithms
by

Alper Guclu
Graduate Research Assistant
480 Town Engineering Building
Department of Civil, Construction and Environmental Engineering
Iowa State University, Ames, IA 50011-3232
Phone: 1-515-294-0223
Fax: 1-515-294-8216
E-mail: alper@iastate.edu

and

Halil Ceylan, Ph.D. (Corresponding Author)


Assistant Professor
482B Town Engineering Building
Department of Civil, Construction and Environmental Engineering
Iowa State University, Ames, IA 50011-3232
Phone: 1-515-294-8051
Fax: 1-515-294-8216
E-mail: hceylan@iastate.edu

(Word Count: Abstract: 202, Text: 4,708, Tables: 1,500, Figures: 1,500, Total: 7,708)

TRB 2007 Annual Meeting CD-ROM Paper revised from original submittal.
Guclu and Ceylan Page 1 of 15

ABSTRACT
The objective of this study was to develop artificial neural network (ANN)-based advanced
backcalculation models as pavement structural analysis tools for the rapid and accurate
prediction of asphalt concrete (AC) overlaid Portland cement concrete (PCC) composite
pavement layer moduli under typical highway loadings. The DIPLOMAT program was used for
solving deflection profiles of composite pavement systems. The DIPLOMAT solutions were
compared with the solutions of ISLAB2000 and ILLI-PAVE pavement analysis programs. ANN-
based backcalculation models trained with the results from the DIPLOMAT solutions have been
found to be practical alternatives for routine pavement evaluation using the falling weight
deflectometer (FWD) deflection data. The trained ANN models in this study were capable of
predicting AC and PCC layer moduli, and the coefficient of subgrade reaction value with low
average absolute errors. A dimensional analysis approach was also adopted by introducing the
dimensional terms of AC modulus over PCC modulus ratio and PCC modulus over coefficient of
subgrade reaction ratio value. Both methods were verified by synthetically generated
DIPLOMAT deflection profiles. ANN-based backcalculation models developed in this study
were also capable of successfully and rapidly (capable of analyzing 100,000 FWD deflection
profiles in one second) predicting the pavement layer moduli from the FWD deflection basins in
real time during field testing. The developed models were successfully validated by results from
the Long-Term Pavement Performance (LTPP) FWD tests conducted on US29, Spartanburg
County, South Carolina.

Key Words: Artificial Neural Networks, Falling Weight Deflectometer, Composite Pavements,
Pavement Layer Backcalculation, Nondestructive Testing and Evaluation, Finite Element
Analysis

TRB 2007 Annual Meeting CD-ROM Paper revised from original submittal.
Guclu and Ceylan Page 2 of 15

INTRODUCTION
Evaluation of pavement layer moduli using the falling weight deflectometer (FWD) has become
the main non-destructive testing technique for the structural evaluation of pavement systems over
the last decade. FWD testing involves measurement of vertical deflections due to dropping of a
mass with known weight from various heights. Sensors located at specific radial distances
monitor the deflection history. The deflections measured at radial distances away from load form
the deflection basin. In order to calculate the pavement structural capacity accurately the
deflection basins should be measured and analyzed accurately. Although there are numerous
methods for evaluating the structural capacity of pavements from deflection basin data, there is
no standard or universally accepted procedure that presently exists [1].

Several pavement layer moduli backcalculation programs have been proposed in the literature.
The AREA method for flexible pavements [2], AREA method for rigid pavements [3-5], ILLI-
SLAB [6], ILLI-BACK [7], best fit algorithm [8-9], ELMOD [10], WESDEF [11],
DIPLOBACK [12], and MODCOMP [13-14] are examples of FWD interpretation programs and
algorithms for rigid, flexible and composite pavements. Backcalculation programs based on
multi-layer elastic layer theory are generally used for AC pavements. For rigid pavements, plate
theory for a slab resting on a Winkler foundation or elastic solid foundation. There is no widely
accepted methodology for AC overlaid PCC type of composite pavements on Winkler
foundation. The backcalculation programs, WESDEF, BISDEF, and ELSDEF, are based on
multi-layer elastic analysis programs, WESLEA, BISAR and ELSYM, respectively. These
programs require the thickness, Poisson’s ratio and a seed modulus as inputs. The forward elastic
layer program iterates the given seed modulus until the given deflections matches with calculated
deflections. Thus, the modulus of pavement layer is highly affected by the seed modulus.
Consequently, experienced engineers are required to use these backcalculation programs [15].
Moreover, elastic layer programs (ELPs) used in asphalt pavement analysis assume linear
elasticity. Pavement geomaterials do not, however, follow a linear type stress-strain behavior
under repeated traffic loading [16, 19]. The ILLI-PAVE [16-18] finite element program which is
commonly used in structural analysis of flexible pavements takes into account nonlinear
geomaterial characterization. Other finite element programs such as ABAQUS, ANSYS, and
DYNA3D are very powerful programs since they can be used in three-dimensional nonlinear
dynamic analysis. Several studies have focused on 3-D finite element modeling of pavements in
last decade [20-22]. Drawbacks associated with these 3-D finite element programs are that
considerable computational resources and time are required for developing a structural model for
each problem.

There are also several finite element based programs exist specifically designed for the analysis
of rigid pavement systems such as ISLAB2000 [23-25]. ISLAB2000 contains many advanced
features that distinguish it from other pavement programs that are based on plate theory.
KENSLABS [26] and WESLIQID [27] are pavement analysis programs for multi-wheel loading
of one- or two-layered medium thick plates resting on a Winkler foundation or elastic solid.
DIPLOMAT [24, 28-29] provides the capability to model pavement layers as plates, springs
and/or elastic layers. DIPLOMAT assumes infinite joints in the horizontal direction. An ANN
based backcalculation procedure was developed for composite pavements by Khazanovich and
Roesler [12] using DIPLOMAT solutions and implemented into program called DIPLOBACK.
DIPLOBACK procedure solutions were agreed with WESDEF solutions [12].

TRB 2007 Annual Meeting CD-ROM Paper revised from original submittal.
Guclu and Ceylan Page 3 of 15

In this paper, the artificial neural networks (ANNs) methodology is applied to rapidly
backcalculate the AC overlaid PCC type composite pavement layer moduli properties. The
pavement structural properties that are of interest in this study are: (1) AC modulus (EAC), (2)
PCC Modulus (EPCC), (3) ks – coefficient of subgrade reaction. To generate a deflection database,
the DIPLOMAT structural analysis program was chosen. DIPLOMAT is chosen specifically for
its capability to analyze pavement layers as plates, elastic, and springs. The results from
DIPLOMAT were compared with those produced by ISLAB2000 and ILLIPAVE. DIPLOMAT
deflection basins were then used to train ANN models for backcalculation of the pavement
structural properties. When compared with the actual DIPLOMAT analysis, the trained ANN
models successfully predicted the pavement layer moduli values, but with several added
advantages.

COMPARISON OF DIPLOMAT WITH OTHER PAVEMENT STRUCTURAL


ANALYSIS PROGRAMS
DIPLOMAT Model
DIPLOMAT is a multi-layered linear elastic structural analysis program for computing pavement
responses (stresses, strains and displacements) subjected to single- or multi-wheel traffic loads
where each load applied over a circular area with a uniform pressure [24]. Each component of
the multi-layered pavement system can be an isotropic, elastic layer, or a plate or a spring layer.
The solution algorithm is based on a generalization of Burmister's layered elastic theory [30-31].
Tensile stresses and downward displacements are assumed to be positive [24]. DIPLOMAT can
accommodate solutions for plate on elastic layer, or elastic layer on plate, or spring models.
Since DIPLOMAT performs analysis using numerical integration methods, it is faster than finite
element programs.

DIPLOMAT vs. ISLAB2000 Comparison


DIPLOMAT and ISLAB2000 solutions were compared to investigate the differences in
pavement deflections. ISLAB2000 is a finite element model based program specifically designed
for analyzing rigid pavement systems. In large part, it is an extension and improvement of the
ILLI-SLAB [6] and ILSL2 [24] programs. ISLAB2000 is a significant improvement over other
finite element programs for the analysis of rigid and composite pavements, enabling users to
analyze a wide range of problems that are not possible with other programs.

Deflection data for the analyzed test pavement section was extracted from both DIPLOMAT and
ISLAB2000 solutions. The layout of a generic AC overlaid PCC test section considered in this
study is given in Figure 1, and the material properties are summarized in Table 1. Primary
consideration in this comparison was how well the DIPLOMAT solutions match with the
ISLAB2000 solutions. Four different slab sizes were considered in ISLAB2000; (1) CP150 -
composite pavement with sizes 3.8 x3.8 m (150 x150 in.), (2) CP300 - composite pavement with
slab sizes 7.6 x 7.6 m (300 × 300 in.), (3) CP600 - composite pavement with slab sizes 15 x 15 m
(600 x 600 in.), and (4) CP900 - composite pavement with slab sizes 23 x 23 m (900 x 900 in.).
All the slabs were loaded at the slab center, and the deflections were extracted at radial distances
of 0, 20, 30, 45, 61, 91, 122, 152 cm (0, 8, 12, 18, 24, 36, 48, and 60 in.) away from the load.
Both programs were analyzed by assuming plate theory for modeling the slab systems. Table 2

TRB 2007 Annual Meeting CD-ROM Paper revised from original submittal.
Guclu and Ceylan Page 4 of 15

summarizes the calculated deflections. The results of this study suggest that the DIPLOMAT
solutions match quite well with ISLAB2000 solutions for large joint spacing which is mainly due
to the infinite slab size assumption in horizontal direction in the DIPLOMAT model.

AC: υ, hAC, EAC hAC

PCC: υ, hPCC, EPCC hPCC

Subgrade: ks

FIGURE 1 Layout of an AC overlaid PCC composite pavement section.

TABLE 1 AC Overlaid PCC Type of Composite Pavement Material Properties

AC Properties PCC Properties Subgrade Properties


EAC = 6.9 GPa (1,000 ksi) EPCC = 27.6 GPa (4,000 ksi) ks = 40.71 kPa/mm (150 psi/in)
hAC = 30 cm (12 in) hPCC = 25 cm (10 in)
Poisson’s ratio = 0.40 Poisson’s ratio = 0.20

TABLE 2 Comparison of ISLAB2000 and DIPLOMAT Solutions

ISLAB2000 Deflections w (mm) DIPLOMAT


FWD Sensor CP150 CP300 CP600 CP900 w (mm)
D0 133 92 85 85 89
D12 128 89 82 82 86
D18 125 87 79 79 83
D24 121 83 76 76 79
D36 113 76 69 68 71
D48 105 67 61 61 63
D60 96 59 53 53 55
D72 89 51 46 45 47

DIPLOMAT vs. ILLI-PAVE Comparison


ILLI-PAVE was developed at the University of Illinois [19] based on the finite element code
used by Duncan et. al. [32] for analyzing highway pavement systems. Since then, numerous
research studies have demonstrated that the ILLI-PAVE model provides a realistic pavement
structural response for highway and airfield pavement systems [16-18]. Recent research studies

TRB 2007 Annual Meeting CD-ROM Paper revised from original submittal.
Guclu and Ceylan Page 5 of 15

at the Federal Aviation Administration’s Center of Excellence established at the University of


Illinois also supported the development of a new, updated version of the program known as the
ILLI-PAVE 2000.

Pavement test section layout and material properties (see Figure 1 and Table 1) used in
ISLAB2000 comparison were used in ILLI-PAVE comparison except that the pavement layers
were assumed to be linear-elastic in both DIPLOMAT and ILLI-PAVE and the layers were
assumed fully bonded in both programs. Table 3 summarizes the results of comparison. ILLI-
PAVE solutions were approximately 15% higher than DIPLOMAT solutions. The deflection
profiles showed the same trend in both cases.

TABLE 3 Comparison of ILLI-PAVE and DIPLOMAT Solutions

FWD Sensor ILLI-PAVE w (mm) DIPLOMAT w (mm)


D0 121.4 106.9
D12 97.7 83.1
D24 92.2 75.9
D36 87.5 68.3
D48 80.9 60.4
D60 73.0 52.4
D72 65.8 44.8

ARTIFICIAL NEURAL NETWORKS (ANNs)


Backpropagation type artificial neural network models were trained in this study with the results
from the DIPLOMAT model and were used as rapid analysis design tools for predicting layer
moduli in AC overlaid PCC composite pavements. Backpropagation ANNs are very powerful
and versatile networks that can be taught a mapping from one data space to another using a
representative set of patterns/examples to be learned. The term “backpropagation network”
actually refers to a multi-layered; feed-forward neural network trained using an error
backpropagation algorithm. The connection weights in the backpropagation ANNs are initially
selected at random. Inputs from the mapping examples are propagated forward through each
layer of the network to emerge as outputs [33]. The errors between those outputs and the correct
answers are then propagated backwards through the network and the connection weights are
individually adjusted to reduce the error. After many examples (training patterns) have been
propagated through the network many times, the mapping function is learned with some
specified error tolerance. This is called supervised learning because the network has to be shown
the correct answers for it to learn. The learning process performed by this algorithm is called
“backpropagation learning” which is mainly an “error minimization technique” [34].

NEURAL NETWORK DESIGN AND TRAINING


ANN analyses were designed with 20,000 DIPLOMAT solutions. The AC overlaid PCC
pavement layout shown in Figure 1 were used. The input ranges are shown in Table 5 below.

TRB 2007 Annual Meeting CD-ROM Paper revised from original submittal.
Guclu and Ceylan Page 6 of 15

TABLE 5 Input Ranges for DIPLOMAT Solutions


AC Modulus PCC Modulus ks hPCC hAC EPCC/ks EAC/EPCC
689 GPa 6,890 GPa 13.5 kPa/mm 152 mm 51 mm
Min 1,023.9 0.008
(100 ksi) (1,000 ksi) (50psi/in) (6 inch) (2 inch)

20,670 82,680 271 508 406


Max 230,674 2.92
(3,000 ksi) (12,000 ksi) (1,000 psi/in) (20 inch) (16 inch)

Based on the theoretical data obtained from DIPLOMAT solutions two main groups of ANN
runs were performed accounting for (1) composite pavement direct method analysis (CDR) runs,
and (2) composite pavement dimensional analysis (CDM) runs. Both CDR and CDM runs had
inputs of deflections at radial distances of 0, 20, 30, 45, 61, 91, 122, 152 cm (0, 8, 12, 18, 24, 36,
48, 60 in.) away from the load which are represented by D0, D8, D12, D18, D24, D36, D48, and D60,
respectively, and pavement layer thickness information to predict the layer moduli of composite
pavement systems. The thickness of AC and PCC were represented as hAC, and hPCC,
respectively. The outputs were: (1) AC modulus (EAC), (2) PCC modulus (EPCC), and (3) ks –
coefficient of subgrade reaction.

A total of 12 ANN runs for CDR model were performed. These runs were categorized into 4
groups depending on the number of deflection data used as follows; (1) CDR-4 with inputs D0,
D12, D24, D36, hAC, hPCC ; (2) CDR-6 with inputs D0, D12, D24, D36, D48, D60, hAC, hPCC; (3)
CDR-7 with inputs D0, D8, D12, D18, D24, D36, D48, D60, hAC, hPCC (Deflections were
according to the FWD Strategic Highway Research Program (SHRP) spacing for the analysis of
LTPP database); (4) CDR-8 with inputs D0, D8, D12, D18, D24, D36, D60, hAC, hPCC. Three
ANN runs were conducted per each group (CDR-4, CDR-6, CDR-7, and CDR-8) to predict three
pavement layer properties; EAC, EPCC, and ks – coefficient of subgrade reaction i.e. CDR4 consists
of CDR-4-EAC, CDR-4-EPCC, and CDR-4-ks. In the same way, CDM runs were prepared as
CDM-4, CDM-6, CDM-7, and CDM-8. Similarly, each group had 3 runs to predict ks- the
coefficient of subgrade, EAC/EPCC ratio, and EPCC/ks ratio. Other inputs were deflection and
thicknesses, except EAC/EPCC runs in which EPCC/ks ratio was added to improve ANN learning.
CDM models were developed to predict EAC and EPCC in stepwise. First the ks value should be
predicted using CDM-ks prediction. Then using the predicted k value and CDM- EPCC/ks run,
EPCC value is predicted. Finally, having known EPCC value, one can predict EAC with CDM-
EAC/EPCC ratio.

ANN dataset consisted of 20,000 DIPLOMAT solutions. This dataset was separated into 18,500
training and 1,500 independent testing sets. ANNs learn the relationship between input
parameters and output variables using the information provided in the training dataset. Then, the
independent 1,500 test data set was used to test how well ANN models have “learned” the
relationship between the input parameters and output variables. A network with two hidden
layers and 60 neurons in each layer was exclusively chosen for the ANN models trained in this
study. Satisfactory results were obtained in the previous studies with these types of networks due
to their ability to better facilitate functional mapping [35-36]. Figure 2 depicts the calculated
mean square error (MSE) at each epoch for training and testing of AC modulus for CDR-8.
MSEs decreased as the networks grew in size with increasing number of epochs. The testing

TRB 2007 Annual Meeting CD-ROM Paper revised from original submittal.
Guclu and Ceylan Page 7 of 15

MSEs were, in general for all models, slightly lower than the training ones. The lowest training
MSEs were in the order of 2.0×10-5 for EAC, 5.2×10-5 for EPCC, 3.6×10-6 for k, 6.0 ×10-5 for
EAC/EPCC ratio, and for 1.3 ×10-4 EPCC/k ratio.

0.025
Training MSE
Testing MSE
0.02
Mean Square Error (MSE)

0.015

0.01

0.005

0
0 1,000 2,000 3,000 4,000 5,000 6,000
Learning Cycles

FIGURE 2 ANN training progress curve for predicting the AC layer moduli.

Average absolute errors for each ANN training are summarized in Table 6 for EAC, EPCC, k,
EAC/EPCC ratio, and EPCC/k ratio predictions. Figure 3 shows the EAC and k prediction
performances for CDR-8 ANN model. Also EAC/EPCC and EPCC/k predictions for CDM-8 are
shown in Figure 4. The outliers at figure 3 and 4 are corresponding to the very thin PCC layers
with very low moduli values.

TABLE 6 Prediction Performance of ANN-Based Backcalculation Models Developed in


this Study
ANN Average Absolute Error (%)
Models EAC EPCC ks EAC/EPCC EPCC/k
CDR-8 0.60 1.10 0.29 - -
CDR-7 0.63 1.11 0.26 - -
CDR-6 0.67 1.00 0.28 - -
CDR-4 0.47 1.16 0.62 - -
CDM-8 - - 0.29 2.37 4.41
CDM-7 - - 0.26 2.59 3.89
CDM-6 - - 0.28 4.05 3.76
CDM-4 - - 0.62 4.1 4.31

TRB 2007 Annual Meeting CD-ROM Paper revised from original submittal.
Guclu and Ceylan Page 8 of 15

ANN ks Prediction (kPa/mm)


25 300
ANN EAC Prediction (GPa)

AAE = 0.6 % AAE = 0.3 %


Training data: 18,500 250 Training data: 18,500
20 Testing data: 1,500 Testing data: 1,500
200
15
Line of Equality Line of Equality
150
10
100

5 Inputs: 8 Def, hAC, hPCC Inputs: 8 Def, hAC, hPCC


50
Output: EAC Output: ks
0 0
0 5 10 15 20 25 0 50 100 150 200 250 300

Given EAC (GPa) Given ks (kPa/mm)


(a) (b )

FIGURE 3 Accuracy of prediction performance for CDR-8 ANN model: (a) EAC, and (b) ks.
ANN EPCC / ks Prediction (x10 )
-3

2.5 250
ANN EAC / E PCC Prediction

AAE = 2.4 % 200 AAE = 4.4 %


2.0
Training data: 18,500 Training data: 18,500
Testing data: 1,500 150 Test data: 1,500
1.5
100
1.0
Line of Equality 50 Line of Equality
0.5
0
Inputs: 8 Def, hAC, hPCC, Inputs: 8 Def, hAC, hPCC
0.0 Output: EAC / EPCC Output: EPCC / ks
0.0 0.5 1.0 1.5 2.0 2.5 0 50 100 150 200 250

Given EAC/EPCC Given EPCC/ks (x 10-3)


(a) ( b)

FIGURE 4 Accuracy of prediction performance for CDM-8 ANN model for: (a) EAC/EPCC,
and (b) EPCC/ks ratio.

TRB 2007 Annual Meeting CD-ROM Paper revised from original submittal.
Guclu and Ceylan Page 9 of 15

VALIDATION
To validate how well the trained ANN models perform using the field FWD data, the LTPP data
site 45-7019 was selected. The FWD data was obtained from the pavement test sections on
US29, Spartanburg County, South Carolina. Original construction date for this site was 1946. It
is an AC overlaid PCC type composite pavement system. FWD deflections were taken from the
“MON_DEFL_DROP_DATA_MT_TN” MS-Access file of LTPP standard data release 20 [36].
LTPP database contains data for the same section for years 1989, 1992, and 1995. The middle
path pavement deflections were chosen for this analysis. LTPP MODCOMP v4.2 backcalculated
values are presented for comparison in Figures 5 and 6. MODCOMP uses elastic layer theory,
embodied in the CHEVRON computer code, as the method of forward calculation within an
iterative approach [1]. Figures 5a and 5b show the ANN CDR-7-AC moduli and LTPP
MODCOMP AC moduli predictions, respectively. Similarly, Figures 6a and 6b show the PCC
moduli predictions for both CDR-7-PCC model and LTPP data set, respectively. As seen in
Figures 5 and 6, EAC predictions are more consistent than the EPCC predictions. Both
MODCOMP and ANN-based EAC values lie within ± 1,378 GPa (200 ksi) range. This range goes
up to ± 6,890 GPa (1,000 ksi) in the case of EPCC. As seen from both AC and PCC prediction
plots, the ANN and MODCOMP predictions, in general, are in good agreement. ANN and
MODCOMP average moduli predictions are given in Figures 5 and 6. The former predictions for
EAC average is 3.2 GPa, and EPCC average is 23.2 GPa, the latter prediction for EAC average is 4.1
GPa, and EPCC average is 25.4 GPa. Few spikes observed in the plots are due to faulty deflection
basins. The scatter for ANN predictions is lower than MODCOMP predictions (see the standard
deviation values presented in Figures 5 and 6) which demonstrates the power of the ANN-based
approach. In addition, ANN moduli predictions for years 1989, 1992, and 1995 are more
consistent compared to the MODCOMP predictions listed in the LTPP database.

50 50
ANN EAC Predictions (GPa)

EAC '89 ‘92 ‘95 EAC '89 ‘92 ‘95


ANN'89 LTPP'89
EAC Predictions (GPa)

AVG 3.5 2.4 3.7 AVG 2.5 3.1 6.9


ANN'92 LTPP'92
LTPP MODCOMP

40 STDEV 2.2 1.7 1.9 40 STDEV 1.9 5.7 5.7


ANN'95 LTPP'95
Mean Value =3.2 GPa Mean Value =4.1 GPa
30 30

20 20

10 10

0 0
0 50 100 150 200 250 300 0 50 100 150 200 250 300

Location (km) Location (km)


(a) ( b)

FIGURE 5 AC layer moduli predictions using: (a) ANN-based models, and (b)
MODCOMP model.

TRB 2007 Annual Meeting CD-ROM Paper revised from original submittal.
Guclu and Ceylan Page 10 of 15

100 100
ANN EPCC Predictions (GPa)
EPCC '89 ‘92 ‘95 EPCC '89 ‘92 ‘95
ANN'89 LTPP'89

EPCC Predictions (GPa)


AVG 25.6 22.4 21.6 AVG 29 25.6 21.4

LTPP MODCOMP
80 ANN'92 80 LTPP'92
STDEV 10.4 9.4 8.3 STDEV 13.9 13.7 9.6
ANN'95 LTPP'95
Mean Value =23.2 GPa Mean Value =25.4 GPa
60 60

40 40

20 20

0 0
0 50 100 150 200 250 300 0 50 100 150 200 250 300

Location (km) Location (km)


(a) (b )

FIGURE 6 PCC layer moduli predictions using: (a) ANN-based models, and (b)
MODCOMP model.

SUMMARY AND CONCLUSIONS


The use of artificial neural networks (ANNs) as pavement analysis tool was demonstrated in this
paper by analyzing asphalt concrete (AC) overlaid Portland cement concrete (PCC) type
composite pavement systems. A total of 12 ANN-based backcalculation models were developed
for predicting each pavement layer moduli using some 20,000 DIPLOMAT model solutions. The
ANN-based models successfully predicted pavement layer moduli values of EAC, EPCC, and k
(modulus of subgrade reaction) with an overall average absolute error (AAE) value of less than
1.5 %. Similarly, ANN-based backcalculation models predicted the EAC/EPCC ratio and EPCC/k
with an AAE of 3.0 % for models in which dimensional analysis was used. It was demonstrated
that ANNs are capable of successfully predicting the pavement layer moduli values using the
LTPP FWD field deflection measurements. Field moduli values were successfully predicted for
the given deflection basins and comparison of the ANN-based predictions with the ones listed in
the LTPP database showed the strength of the ANN-based backcalculation approach. Such ANN
models are invaluable tools for pavement engineers for evaluating the structural condition of
composite pavement systems.

The adoption of ANN-based approach also resulted in both a drastic reduction in computation
time and a simplification of the complicated traditional layer backcalculation approaches. Rapid
prediction ability of the ANN models - capable of analyzing 100,000 FWD deflection profiles in
one second - provide a tremendous advantage to the pavement engineers by allowing them to
nondestructively assess the condition of the transportation infrastructure systems in real time
while the FWD testing takes place in the field. Elimination of selecting seed layer moduli with
the integration of ANN-based direct backcalculation approach can be invaluable for the state and
federal agencies for rapidly analyzing large number of composite pavement deflection basins
needed for routine pavement evaluation for both project specific and network level FWD testing.

TRB 2007 Annual Meeting CD-ROM Paper revised from original submittal.
Guclu and Ceylan Page 11 of 15

ACKNOWLEDGEMENTS
The authors gratefully acknowledge the Iowa Department of Transportation (IA-DOT) for
sponsoring this study. The contents of this paper reflect the views of the authors who are
responsible for the facts and accuracy of the data presented within. The contents do not
necessarily reflect the official views and policies of the IA-DOT. This paper does not constitute a
standard, specification, or regulation.

REFERENCES
(1) PCS/Law Engineering, SHRP's Layer Moduli Back-Calculation Procedure: Software
Selection, SHRP-P651, Washington, DC: Strategic Highway Research Program, National
Academy of Science, 1993.

(2) Hoffman, M. S., and M. R. Thompson. Backcalculating Nonlinear Resilient Moduli from
Deflection Data. Transportation Research Record 852, TRB, National Research Council,
Washington, D.C., 1982, pp. 42-51.

(3) Ioannides, A. M., E. J. Barenberg, and J. A. Lary. Interpretation of Falling Weight


Deflectometer Results Using Principals of Dimensional Analysis. Proceedings, 4th
International Conference on Concrete Pavement Design and Rehabilitation, Purdue
University, 1989, pp.231-247.

(4) Ioannides, A. M. Dimensional Analysis in NDT Rigid Pavement Evaluation. Journal of


Transportation Engineering, Vol.116, No. 1, 1990, pp.23-36.

(5) Barenberg, E. J., and K. A. Petros. Evaluation of Concrete Pavements Using NDT Results.
Project IHR-512, University of Illinois at Urbana-Champaign and Illinois Department of
Transportation, Report No. UILU-ENG-91-2006, 1991.

(6) Foxworthy, P. T., and M. I. Darter. ILLI-SLAB and FWD Deflection Basins for
Characterization of Rigid Pavements. Nondestructive Testing of Pavements and
Backcalculation of Moduli. American Society for Testing and Materials, 1989, pp. 368-386.

(7) Ioannides, A.M.. Concrete Pavement Backcalculation Using ILLI-BACK 3.0,


Nondestructive Testing of Pavements and Backcalculation of Moduli. American Society for
Testing and Materials, Vol. 2, 1994, pp.103-124.

(8) Hall, K. T., M. I. Darter, T. Hoerner, and L. Khazanovich. LTPP Data Analysis – Phase I:
Validation of Guidelines for K Value Selection and Concrete Pavement Performance
Prediction. Interim Report prepared for FHWA, ERES Consultants, Champaign, IL, 1996.

(9) Smith, K. D., M. J. Wade, D. G. Peshkin, L. Khazanovich, H. T. Yu, and M. I. Dater.


Performance of Concrete Pavements, Volume II – Evaluation of In-Service Concrete
Pavements. Report No. FHWA-RD-95-110, ERES Consultants, Champaign, IL, 1996.

TRB 2007 Annual Meeting CD-ROM Paper revised from original submittal.
Guclu and Ceylan Page 12 of 15

(10) Ullidtz, P., “ELCON: Evaluation of Layer Moduli and Overlay Design (ELMOD for
Concrete)”, Dynatest Engineering A/S, Denmark, 1987.

(11) Van Cauwelaert, F. J., Alexander, D. R., White, T. D., and W. R. Barker. Multilayer
elastic program for elastic program for backcalculating layer moduli in pavement evaluation.
Nondestructive Testing of Pavements and Backcalculation of Moduli, ASTM STP 1026,
Philadelphia, USA, 1989.

(12) Khazanovich, L. and Roesler, J.. DIPLOBACK: Neural Network based Backcalculation
Program for Composite Pavements, Transportation Research Record No. 1570, TRB,
National Research Council, Washington, D.C., 1997, pp. 143-150.

(13) Irwin, L. H., Szenbenyi, T. User’s Guide to MODCOMP3 Version 3.2, CLRP Report
Number 91-4, Cornell University, Local Roads Program, Ithaca, NY,1991.

(14) Irwin, L. H., Instructional Guide for Back-Calculation and the Use of MODCOMP,
CLRP Publication No. 94-10, Cornell University, Local Roads Program, Ithaca, NY, 1994.

(15) Lytton, R.L. “Backcalculation of Pavement Layer Properties” Nondestructive Testing of


Pavements and Backcalculation of Moduli” ASTM Testing and Materials, Philadelphia, pp.
7-38., 1989.

(16) Brown, S.F. and Pappin, J.W. Analysis of Pavements with Granular Bases.
Transportation Research Record 810, TRB, National Research Council, Washington, D.C.,
pp. 17-23. 1981.

(17) Thompson, M. R. and R. P. Elliot ILLI-PAVE Based Response Algorithms for Design of
Conventional Flexible Pavements. Transportation Research Record 1043, TRB, National
Research Council, Washington, D.C. 1985.

(18) Garg, N. E., Tutumluer, E., and M. R. Thompson. Structural Modelling Concepts for the
Design of Airport Pavements for Heavy Aircraft. Proceedings of the Fifth International
Conference on the Bearing Capacity of Roads and Airfields, Trondheim, Norway, 1998.

(19) Raad, L. and J. L. Figueroa Load Response of Transportation Support Systems.


Transportation Engineering Journal, ASCE, Vol 16, No. TE1., 1980.

(20) Mallela, J., and K.P. George. Three-Dimensional Dynamic Response Model for Rigid
Pavements. In Transportation Research Record: Journal of the Transportation Research
Board, No. 1448, TRB, Washington, D.C., 1994.

(21) Darter, M.I., K.T. Hall, and C. Kuo. Support Under Portland Cement Concrete
Pavements. NCHRP Report 372. Washington, DC: National Cooperative Highway Research
Program, 1995.

TRB 2007 Annual Meeting CD-ROM Paper revised from original submittal.
Guclu and Ceylan Page 13 of 15

(22) Kennedy J.C. Material Nonlinear and Time-Dependent Effects on Pavement Design for
Heavyweight, Multi-Wheel Vehicles. Proceedings of the First International symposium on
3D Finite Element for Pavement Analysis and Design, 1998.

(23) Tabatabaie, A.M., and E.J. Barenberg. Finite Element Analysis of Jointed or Cracked
Concrete Pavements. In Transportation Research Record: No. 671, TRB, 1978, pp. 11-18.

(24) Khazanovich, L. Structural Analysis of Multi-Layered Concrete Pavement Systems, Ph.D.


dissertation, University of Illinois, Illinois, USA, 1994.

(25) Khazanovich, L., Yu, H.T., Rao, S., Galasova, K., Shats, E., and Jones, R. ISLAB2000 -
Finite Element Analysis Program for Rigid and Composite Pavements, User’s Guide, ERES
Consultants, A Division of Applied Research Associates, Champaign, Illinois, 2000.

(26) Huang, Y.H. A Computer Package for Structural Analysis of Concrete Pavements.
Proceedings, 3rd International Conference on Concrete Pavement Design and Rehabilitation,
Purdue University, 1985, pp.295-307.

(27) Chou, Y. T. Structural Analysis Computer Programs for Rigid Multicomponent


Pavement Structures with Discontinuities - WESLIQID and WESLAYER. Technical Report
GL-81-6, U.S. Army Engineer Waterways Experiment Station, May 1981.

(28) Ioannides, A.M. and Khazanovich, L., “Analytical and Numerical Methods for Multi-
Layered Concrete Pavements” Proceedings, Third International Workshop on the Design and
Rehabilitation of Concrete Pavements, Krumbach, Austria, October, 1994.

(29) Khazanovich, L., and Ioannides, A.M. DIPLOMAT: Analysis Program for Bituminous
and Concrete Pavements. In Transportation Research Records: Journal of the Transportation
Research Board, No. 1482, 1995, pp. 52-60.

(30) Burmister, D.M. “The Theory of Stresses and Displacements in Layered Systems and
Application to the Design of Airport Runways “Proceedings, Highway Research Board, Vol.
23, National Research Council, Washington D.C, 1943, pp. 126-144.

(31) Burmister, D.M. “The General Theory of Stresses and Displacements in Layered
Systems, “Journal of Applied Physics, 1945, Vol. 16, No. 2, pp. 89-96; No.3 pp. 126-127;
No.5, pp. 296-302.

(32) Duncan, J.M., Monismith, C.L., and Wilson, E.L. “Finite element analyses of
pavements.” In Highway Research Record No. 228, HRB, National Research Council,
Washington D.C., 1968.

(33) Meier, R. W. and Rix, G. J., "Backcalculation of Flexible Pavement Moduli from
Dynamic Deflection Basins Using Artificial Neural Networks," Transportation Research
Record 1473, TRB, National Research Council, Washington, D. C., 1995

TRB 2007 Annual Meeting CD-ROM Paper revised from original submittal.
Guclu and Ceylan Page 14 of 15

(34) Haykin, S. Neural networks: A comprehensive foundation. Prentice-Hall, Inc., NJ, USA.
1999.

(35) Ceylan, H. Analysis and Design of Concrete Pavement Systems Using Artificial Neural
Networks, Ph.D. Dissertation, University of Illinois, Urbana-Champaign, IL, December,
2002.

(36) Ceylan, H., Guclu, A., Tutumluer, E., and Thompson, M.R. Backcalculation of Full-
Depth Asphalt Pavement Layer Moduli Considering Nonlinear Stress-Dependent Subgrade
Behavior. The International Journal of Pavement Engineering, Vol. 6, No 3, 2005, pp. 171–
182.

(37) LTPP Standard Data Release 20, Long-Term Pavement Performance (LTPP) program,
Federal Highway Administration, Washington DC, 2005.

TRB 2007 Annual Meeting CD-ROM Paper revised from original submittal.

Das könnte Ihnen auch gefallen