You are on page 1of 7

Improved Method for Evaluating the

Pavement Structural Number with Falling


Weight Deflectometer Deflections
Moo Yeon Kim, Dae Young Kim, and Michael R. Murphy

The pavement structural number (SN) is used in various applications SN eff


worldwide. One application is the structural condition index. The struc-
structural condition index = (1)
SN req
tural condition index was developed for the Texas Department of Trans-
portation (DOT) to support maintenance and rehabilitation decisions at
where SNeff is the effective SN and SNreq is the required SN. Because
the network level and is calculated as the ratio of two SN values (effective
SN values are sensitive enough to pavement deterioration, Zhang et
SN/required SN). A direct method of determining the SN from falling
al. concluded that an SN value could be a good indicator of the
weight deflectometer (FWD) testing has been used to evaluate the effec-
effective structural condition of a pavement.
tive SN. This study sought to increase the accuracy of the structural con-
The structural condition index can be interpreted as follows: if the
dition index by improving the method of determining the effective SN
value is 1 or greater than 1, then the effective pavement condition
from FWD deflection data. A new equation was developed by modifying
is better than the required pavement condition. In other words, the
the equation that was previously used, and new coefficients for four dif-
pavement is in sound condition for the estimated future equivalent
ferent flexible pavement types were used. The new equation was improved
single-axle loads. If the value of the structural condition index is less
by the addition of new variables for considering the depth to a rigid layer
than 1, the pavement will need a certain kind of treatment. Peddib-
and by the use of a large database consisting of hypothetical pavement
hotla et al. conducted a threshold analysis to establish guidelines for
structures and responses. This effort not only benefits the Texas DOT by
selecting the appropriate category of maintenance and rehabilita-
allowing increased accuracy in implementing the structural condition
tion treatment, such as preventive maintenance, light rehabilitation,
index but also helps other agencies improve their evaluation of the SN by
medium rehabilitation, or heavy rehabilitation (2).
FWD deflection at the network level. In this study, only surface-treated
The Austin District of the Texas DOT has adopted the structural
pavement and asphalt concrete pavement were evaluated; portland
condition index as a pavement evaluation tool at the network level.
cement concrete pavement was not addressed.
To provide a convenient way for Texas DOT engineers to estimate the
structural condition index, the Center for Transportation Research at
To preserve or improve the surface condition of pavements, the the University of Texas at Austin developed an automated program
Texas Department of Transportation (DOT) has employed measures that uses a Microsoft Excel macro to calculate the index. In the cal-
such as extensive seal coats, thin overlays, and other types of surface culation of the structural condition index, the required SN can be
treatments. Although these treatments improve the surface condition obtained from the AASHTO design method, which is based on esti-
temporarily, the general state of the surface condition deteriorates mated equivalent single-axle loads for the subsequent 20 years. The
continuously because of the structural deformation of pavement effective SN is calculated by determining the SN from FWD testing
systems. Accordingly, accurately characterizing a pavement’s struc- as proposed by Rohde (3); this method depends on FWD deflections
tural condition is essential to choosing an appropriate treatment to and the total thickness of the pavement layers. Although the struc-
effectively help prevent or delay pavement deterioration. tural condition index is a sound concept, its accuracy relies entirely
Zhang et al. developed a structural condition index for the Texas on Rohde’s equation. Therefore, an examination of the equation and
DOT to assist with maintenance and rehabilitation decisions (1). an endeavor to improve its accuracy are desirable.
This index was based on currently available information such as This paper presents a new equation modified from Rohde’s original
falling weight deflectometer (FWD) deflection data and data from analysis (3). The new equation was improved by the addition of new
the Texas DOT’s Pavement Management Information System (PMIS). variables for considering the depth to a rigid layer and by the use of
The structural condition index is calculated as a ratio of the effective a large database consisting of hypothetical pavement structures and
and the required pavement structural numbers (SNs): responses.

M. Y. Kim and M. R. Murphy, Center for Transportation Research, University
of Texas at Austin, 1616 Guadalupe Street, Suite 4.202, Austin, TX 78701. Background
D. Y. Kim, Department of Architectural Engineering, Dong-Eui University, 176
Eomgwangno Busanjin-Gu, Pusan 614-714, South Korea. Corresponding author: Equation for Estimating SN Directly
M. Y. Kim, mooyeon@utexas.edu. from FWD Deflections
Transportation Research Record: Journal of the Transportation Research Board,
No. 2366, Transportation Research Board of the National Academies, Washington,
Rohde presented the equation for determining the SN directly from
D.C., 2013, pp. 120–126. the FWD deflection basin (3). His methodology was based on the
DOI: 10.3141/2366-14 two-thirds rule proposed by Irwin (4), which is illustrated in Figure 1.

120
Kim, Kim, and Murphy 121

FWD Load Plate


FWD Load FWD Sensors

34°
Hp

1.5 Hp 17.75” (450mm)

SIS

SIP
FWD Deflection Bowl

FIGURE 1   Stress distribution and measured deflection bowl beneath FWD load (3).

This law states that about 95% of deflections originate below a line instead, viscous, viscoelastic, or plastic behaviors accompany the
angling 34° from the horizontal surface. In this approach, deflection elastic response.
beyond 1.5 times the total thickness of the pavement layer (Hp) is This method’s advantage is that determination of the thickness
primarily affected by the subgrade. On the basis of this assumption, of each pavement layer is not necessary. The total thickness used
the difference between the peak deflection and the deflection at in the equation is relatively easy to obtain as compared with the
1.5 Hp strongly correlates with the pavement structure. Rohde’s individual layer thickness, which is one reason that the calculation
equation is as follows: of the structural condition index uses this approach. However, this
method has a few drawbacks. First, the depth to a rigid layer is not
SN = k1SIP k2 Hp k3 (2) taken into account. For that reason, with the same value of SN cal-
culated from Equation 3, it is possible to have a rather large range
where of SNs estimated by this method. For instance, it is indicated that
data points are vertically aligned at the same SN as calculated
SN = structural number (in.);
by the AASHTO equation (x-axis) in Figure 2. The magnitude of
SIP = structural index of pavement, D0 − D1.5Hp (µm);
this dispersion is up to approximately two units in a typical SN
Di = surface deflection (in. • 10−3) normalized to 9,000-lb
range. The amount of deviation can be reduced by employing the
(40-kN) load at offset i (ft; 1 mil = 25.4 µm);
concept of the depth to a rigid layer, which increases the accuracy
Hp = total thickness of pavement (mm); and
of SN evaluation.
k1, k2, k3 = coefficients.
Another downside of this method is the number of hypothetical
To obtain Equation 2, Rohde created 7,776 hypothetical pavement pavement structures involved. Rohde’s study analyzed 7,776 struc-
structures with various combinations of stiffness–thickness. The SN tures of various combinations of stiffness and thickness. Although the
was calculated with Equation 3, which was taken from the AASHTO range of pavement combinations led to a reasonable result, it was not
Guide for Design of Pavement Structures (5, 6): enough to represent most cases of flexible pavement in Texas. There-
fore, the present study employed a large, Texas-specific database
n
 E 3
1
(NETFWD) to improve the method.
SN = ∑ hi ag  i  (3)
i =1  Eg 
NETFWD Method
where
ag = layer coefficients of standard materials (AASHO road test), NETFWD is a structural evaluation methodology based on a data-
Ei = in situ layer modulus, base search methodology developed by Murphy to take advantage
Eg = resilient modulus of standard materials (AASHO road test), of existing information in the PMIS database as well as to com-
hi = layer thickness (in.), and pensate for the lack of information on layer thickness in the PMIS
SN = structural number (units of hi).
Equation 2 and its coefficients were then evaluated to find the best TABLE 1   Coefficients for SN versus SIP Relationships (3)
relationship between the SN values calculated from the two differ-
Surface Type k1 k2 k3 r2 n
ent methods. As shown in Table 1, fairly good correlation was found
for both types of flexible pavement—surface seals and asphalt Surface seals 0.1165 −0.3248 0.8241 .984 1,944
concrete pavement. However, Rohde noted that the relationship Asphalt concrete 0.4728 −0.4810 0.7581 .957 5,832
is purely theoretical and is based on layer elastic theory (3). Real
pavement structures are not likely to have pure elastic behaviors; Note: r2 = coefficient of determination; n = sample size.
122 Transportation Research Record 2366

FIGURE 2   Correlation of SN obtained by using Equation 3 versus Equation 2 (3).

(7). Murphy built an extensive database of hypothetical pavement divided by ranges of surface thickness, predict the depth to a rigid
structures and responses based on surveys that investigated the typi- layer as a function of zero deflection point (r0) and FWD deflection
cal ranges of layer thickness and modulus in Texas. This database basin parameters, including the surface curvature index (SCI), the base
contains FWD deflections, stresses, and strains computed by the damage index (BDI), and the base curvature index (BCI). Rohde and
BISAR program, which is based on linear elastic layered theory. Smith verified this equation with field data, and later Mahoney et al.
FWD deflection-matching criteria can be used to search the data- evaluated these equations (9), developing a method adopted for use in
base to find a pool of candidate solutions for pavement structures, the Washington State DOT’s backcalculation program EVERCALC.
stresses, and strains. The database uses three layers—(a) surface, The proposed equations are as follows:
(b) base, and (c) subgrade and rigid bottom—as hypothetical struc-
tures. Each layer has various ranges of thickness and moduli; there- 1
= 0.0362 − 0.3242r0 + 10.2717r 02 − 23.6609r 03 − 0.0037BCI
fore, the database yields a large number of combinations, as shown B
in Table 2. NETFWD consists of four different Texas DOT PMIS
(4)
pavement types (Types 4, 5, 6, and 10) categorized by total surface
layer thickness. In total, 15,554,347 pavement combinations exist
1
in the NETFWD database. = 0.0065 + 0.1652r0 + 5.42898r 02 − 11.0026r 03 + 0.004BDI
B
(5)
Method for Estimating the Depth to a Rigid Layer
1
Rohde and Smith developed four regression equations for estimating = 0.0413 + 0.9929r0 − 0.0012SCI + 0.0063BDI
B
the depth to a rigid layer with FWD data; these equations were for
use in the MODULUS backcalculation program (8). These authors − 0.0778 log ( BCI ) (6)
used the Boussinesq equations and the relationship between surface
deflection and the inverse of radial offset to estimate the depth at which 1
= 0.0409 + 0.5669r0 + 3.0137r02 + 0.0033BDI
zero deflection would occur, as shown in Figure 3. The four equations, B
− 0.0665 log ( BCI ) (7)

TABLE 2   Categorization and Number of NETFWD Data where


Total Surface Number of B = depth to a rigid layer (ft),
PMIS Class Pavement Type Thickness (in.) Structures r0 = 1/r intercept by extrapolating the steepest section of the 1/r
versus deflection curve,
4 ACP >5.5 8,261,212
SCI = D0 − D1,
5 ACP   2.5–5.5 4,459,908
BDI = D1 − D2, and
6 ACP <2.5 2,041,292 BCI = D2 − D3.
10 Surface treatment <2.5 791,935
Total — — 15,554,347 Equations 4 through 7 were not directly used in the present study
because the authors questioned the efficiency of developing a regres-
Note: ACP = asphalt concrete pavement; — = not applicable. sion equation on the basis of another regression equation. However,
Kim, Kim, and Murphy 123

FIGURE 3   Deflection versus inverse of offset (1/r) for several hypothetical pavement
structures (8).

the zero deflection point and FWD deflection basin parameters were Determination of Variables for New Equation
adopted to account for the depth to a rigid layer.
Rohde’s method of estimating the SN from FWD deflections (Equa-
tion 2) has variables such as SIP and Hp (3). The authors considered
Methodology four additional variables, including the zero deflection point (r0) and
three FWD deflection basin parameters (SCI, BDI, and BCI) to take
Reduction of NETFWD Data into account the depth to a rigid layer.
The SIP plays a critical role in evaluating the SN with Rohde’s
A reduced set of data from the original NETFWD database was approach. To obtain the SIP, the surface deflection at the offset of
used because dealing with a large database causes difficulties such 1.5 Hp has to be interpolated by using the other known deflections.
as regression, interpolation, and other calculation-based errors. A Rohde’s study proposed use of the Lagrange polynomial interpolation
Microsoft Excel spreadsheet accommodates approximately 1 mil- with n = 3, as follows (3):
lion rows, which provides a target number for reduction of the data
set. Therefore, the original data set was divided into four data sets ( Rx − RB )( Rx − RC ) ( R + RA )( Rx − RC )
Dx = DA + x DB
according to PMIS pavement type. The data set was further reduced ( RA − RB ) ( RA − RC ) ( RB − RA ) ( RB − RC )
by omitting rare or uncommon cases of pavement structural com-
( Rx − RA )( Rx − RB )
binations in Texas. Because the developer of the NETFWD data- + DC (8)
base considered extreme cases to meet all possible future situations, ( RC − RA )( RC − RB )
some rare combinations were generated that do not represent typi-
cal pavement structures in Texas. In addition, the intervals of layer where
thickness and modulus were expanded to avoid redundant calcu- Dx = deflection at offset of Rx;
lations. As a result of this procedure, the data extracted from the Di = deflection at sensor i;
original NETFWD data were reduced to the totals shown in Table 3. Ri = offset of sensor i;
In total, 1,194,514 combinations—a figure that provides sufficient i = A, B, C, being three closest sensors to point x; and
pavement structures for evaluation—were prepared for this study. x = point for which deflection is determined.

TABLE 3   Reduced NETFWD Data

Surface Base Subgrade


Number of
PMIS Class Structures Thickness (in.) Modulus (ksi) Thickness (in.) Modulus (ksi) Thickness (in.) Modulus (ksi)

Type 4 567,159 6–12 20–1,250 4–18 15–500 60–720 4–45


Type 5 390,663 2.5–5.5 50–1,250 6–18 10–500 60–720 4–45
Type 6 161,188 1.5–2 50–1,250 4–18  5–500 60–720 4–45
Type 10 75,504 1–2 50–150 4–18  4–500 60–720 4–45

Note: 1 ksi = 6.89 MPa.


124 Transportation Research Record 2366

Deflection (mils)

W0
100

80

60

40

Interpolating Point
20 W12
W24
0 Sensor Offset (in.)
5 10 15 20 25 30

−20

FIGURE 4   Example of wrong interpolation of FWD deflection basin.

The authors noticed that, in some cases, the suggested interpolation where
method yielded unrealistic values, such as negative deflection. Fig-
SN = structural number (in.),
ure 4, for example, shows one of the cases in which the interpolated
SIP = D0 − D1.5Hp (mils),
value was negative in the FWD deflection versus the offset plot.
Hp = total pavement thickness (in.), and
Because of this observation, the linear interpolation was used in
k1, k2, k3, k4, k5, k6, k7, k8 = coefficients.
the evaluation of the SIP, which should always have an adequate
approximation. The coefficients for the relationship are given in Table 4.
The zero deflection point r0 can be obtained by comparing the Figure 5 demonstrates the effect of using the NETFWD data and
FWD deflection to the inverse of the offset plot and finding the inter- the new equation (Equation 9). For each of the four PMIS pavement
cept by extrapolating the steepest section. As shown in Figure 3, classifications, the figure compares the results from three different
the value of r0 decreases proportionally to the depth to a rigid layer. approaches:
Thus, r0 can be a useful variable in calculating the depth to a rigid
layer. The three FWD deflection basin parameters were used as Approach 1 (left graph). Correlation between the SN calculated
variables because Rohde’s equations for estimating the depth to a from Rohde’s equation (Equation 2) with its original coefficients
rigid layer (Equations 4 through 7) used those three variables, and he and the SN calculated from the AASHTO equation (Equation 3).
found a good correlation between the depth to a rigid layer and the Approach 2 (middle graph). Correlation between the SN calcu-
FWD deflection basin parameters. lated from Equation 2 with the modified coefficients calibrated by
NETFWD and the SN calculated from Equation 3.
Approach 3 (right graph). Correlation between the SN derived from
Development of New Equation the new equation (Equation 9) and the SN calculated from Equation 3.

With the determined variables, several equation formats were cho- For Type 10 (surface treatment pavement), Approach 1 shows that
sen as candidates. After several trials to determine the appropriate Rohde’s equation (Equation 2) generally underestimates the SN. In
form of equation among the candidates, the authors found a non- contrast, Approach 2 clearly corrects such underestimation by adopt-
linear regression equation with eight coefficients for each PMIS ing the calibrated coefficients, as shown in Table 5. Moreover, the
pavement type. deviation of Approach 3 is visibly smaller than that of Approaches
1 and 2. This result becomes clearer when values of the coefficient
SN = k1SIP k2 Hp k3 + k4 (1 + r0 ) 5 + k6SCI k7 BDI k8
k
(9) of determination in Tables 4 and 5 are compared. In Approach 3,

TABLE 4   Coefficients for Equation 9

PMIS Class k1 k2 k3 k4 k5 k6 k7 k8 r2

Type 4 1.5136 −0.1749 0.5718 −1.7876 1.9035 1.2479 0.1411 −0.8383 .960
Type 5 0.9712 −0.3129 0.7521 −0.3422 2.9334 0.0657 0.8854 −2.5504 .970
Type 6 0.2797 −0.3770 1.1606 −0.0149 3.4038 1.2251 −0.8397 0.6901 .989
Type 10 0.2296 −0.4112 1.2626  0.1629 0.7919 1.5146 −1.4247 1.0960 .988
(a)

(b)

(c)

(d)

FIGURE 5   Comparison of results obtained from three approaches for four types of pavement: (a) Type 10, (b) Type 6,
(c) Type 5, and (d) Type 4.
126 Transportation Research Record 2366

TABLE 5   New Coefficients for Rohde’s Equation

Original Coefficient New Coefficient

PMIS Class k1 k2 k3 r2 k1 k2 k3 r2

Type 4 0.4728 −0.4810 0.7581 .841 0.0875 −0.2926 0.8860 .890


Type 5 0.4728 −0.4810 0.7581 .929 0.1361 −0.3719 0.8752 .962
Type 6 0.4728 −0.4810 0.7581 .940 0.1070 −0.3715 0.9179 .980
Type 10 0.1165 −0.3248 0.8241 .977 0.0811 −0.3994 0.9904 .980

the new equation provides the highest coefficient of determina- The new equation proposed here can be used to determine a reason-
tion. A tendency to increase the dispersion of estimated values as SN able SN value solely from FWD deflection data and total pavement
increases is evident in the Approach 3 graph. However, in practice, thickness; no other information is required. This effort not only ben-
typical Type 10 pavement rarely has an SN value above 4. Thus, efits the Texas DOT by allowing increased accuracy in implementing
such deviations are tolerable. the structural condition index, but also helps other agencies improve
For Types 5 and 6—asphalt concrete pavements with a surface their evaluation of the SN by FWD deflection at the network level.
thickness of between 2.5 in. and 5.5 in. (Type 5) and less than 2.5 in.
(Type 6)—the correlations seem fairly good for all approaches,
although the data points are more dispersed as the surface layer thick- Acknowledgment
ness increases. Nonetheless, Approach 1 illustrates moderately good
correlation, Approach 2 shows a better relationship, and the new This study was conducted as part of a research project sponsored by
method provides even better results in terms of dispersion. In other the Texas Department of Transportation.
words, the use of the NETFWD database increased the accuracy
of Rohde’s equation, and the use of the new equation together with
NETFWD resulted in significant improvement. References
Type 4, the thickest pavement classification (>5.5 in.), shows
a very wide band of data. Rohde’s approach (Approach 1) shows 1. Zhang, Z., G. Claros, L. Manuel, and I. Damnjanovic. Development
of Structural Condition Index to Support Pavement Maintenance and
the worst relationship. Approach 2 also displays a wide and split Rehabilitation Decisions at Network Level. In Transportation Research
band, as does Approach 1. That tendency might result from a lack in Record: Journal of the Transportation Research Board, No. 1827, Trans-
Rohde’s study of data on pavements with a surface thickness greater portation Research Board of the National Academies, Washington, D.C.,
than 8 in. The new method found a better relationship for this type 2003, pp. 10–17.
of pavement. However, this correlation should be used with care 2. Peddibhotla, S. S. S., M. R. Murphy, and Z. Zhang. Validation and Imple-
mentation of the Structural Condition Index (SCI) for Network-level
because of its wide range. Pavement Evaluation. FHWA/TX-11/5-4322-01-1. Texas Department of
All in all, the proposed new method provides improved results for Transportation, Austin, 2010.
evaluating SN with FWD surface deflection data. 3. Rohde, G. T. Determining Pavement Structural Number from FWD Test-
ing. In Transportation Research Record 1448, TRB, National Research
Council, Washington, D.C., 1994, pp. 61–68.
4. Irwin, L. H. User’s Guide to Modcomp2, Version 2.1. Local Roads Program,
Summary and Conclusions
Cornell University, Ithaca, N.Y., 1983.
5. Guide for Design of Pavement Structures. AASHTO, Washington, D.C.,
This paper described the improvement of a method for evaluating 1986.
pavement SN by using FWD deflections. The use of NETFWD, a 6. Guide for Design of Pavement Structures. AASHTO, Washington, D.C.,
large database of hypothetical pavement structures, provided strong 1993.
theoretical correlation between the SN evaluated by the new method 7. Murphy, M. R. A Mechanistic-Empirical Approach to Characterizing
and the SN from the AASHTO approach. By employing the concept Subgrade Support and Pavement Structural Condition for Network-level
Applications. PhD dissertation. University of Texas at Austin, 1998.
of the depth to a rigid layer, it was possible to improve the SN evalu- 8. Rohde, G. T., and R. E. Smith. Determining Depth to Apparent Stiff
ation method. Moreover, the new method provides coefficients for Layer from FWD Data. Research Report FHWA/TX-91-1159-1. Texas
four different PMIS pavement classifications as opposed to the orig- Department of Transportation, Austin, 1991.
inal approach (3), which has coefficients for two pavement types. In 9. Mahoney, J. P., B. C. Winters, N. C. Jackson, and L. M. Pierce. Some
this respect, the new method demonstrates significant improvement. Observations About Backcalculation and Use of a Stiff Layer Condi-
tion. In Transportation Research Record 1384, TRB, National Research
This improvement increases the accuracy of evaluation of the struc- Council, Washington, D.C., 1993, pp. 8–14.
tural condition index and thus enhances the accuracy of identifying
structurally sound pavements and those that require treatment at the
network level. The Pavement Management Systems Committee peer-reviewed this paper.

You might also like