Professional Documents
Culture Documents
Over the past decade the backcalculation of layer moduli, primarily from Falling Weight
Deflection (FWD) measurements, has become a primary component of pavement evaluation and
rehabilitation design methods. Despite the many limitations of layered elastic theory in
modelling real pavement behaviour, almost all backcalculation programs utilize layered elastic
theory or some derivative thereof, and popular examples include MODULUS (1), EVERCALC
(2), and ELMOD (3). The input typically required for such programs consists of measured
deflections, layer thicknesses, seed moduli or moduli ranges, and the applied load or pressure.
Whilst a substantial amount of research has been done on the development and validation
of backcalculation programs, comparatively little work has been done to quantify the influence
There are two generic types of error involved: systematic and random. Systematic errors
can be limited or even eliminated through proper measurement and analysis procedures (such as
the Strategic Highway Research Program, SHRP, calibration procedure for FWDs). However
random errors in input data are unavoidable, due to the nature of road building equipment,
materials and processes. The inherent variability in material properties and the relatively large
tolerances allowed in construction therefore stand in direct contrast to the complexity and
notional accuracy of the models that are used to characterize pavement behaviour.
F.J. Jooste, Division of Roads and Transport Technology, CSIR, P.O. Box 395, Pretoria, 0001, South Africa,
telephone: +027-12-841-2224, facsimile: +27-12-841-3095
S.V. Kekwick, Division of Roads and Transport Technology, CSIR, P.O. Box 395, Pretoria, 0001, South Africa,
telephone: +027-12-841-3080/3901, facsimile: +27-12-841-3095
M. Muthen, Division of Roads and Transport Technology, CSIR, P.O. Box 395, Pretoria, 0001, South Africa,
telephone: +027-12-841-2115, facsimile: +27-12-841-3095
Jooste, Kekwick and Muthen 2
of layer thickness. At project level, the accuracy of estimated layer thicknesses can be improved
by increasing the frequency of coring and test pit positions. In many instances, however, cost
limitations or the mere scope of the project preclude such an accurate determination. In such
cases thicknesses must then be derived from secondary sources of information, such as design
or as-built plans, with this approach being typical for pavement analysis carried out at network
level. The information is then used to define uniform pavement sections for which the layer
A factor that is seldom taken into account during backcalculation is the variation in layer
thicknesses due to construction tolerances. Even within an acceptably uniform pavement section,
an inevitable variation in layer thickness will exist. Provided that this variation is within
specified tolerances, it is not a construction error, yet it may have a significant influence on the
backcalculation process.
Although the influence of layer thickness variation on backcalculated moduli has been
investigated by other researchers (4,5), such work normally deals with incorrect estimates of
layer thicknesses. This paper, on the other hand, deals with the effects of inherent, acceptable
ANALYSIS APPROACH
Since most backcalculation programs make use of layered elastic theory, this study uses the same
basis and is, in effect, an analysis of the sensitivity of layered elastic theory to small variations
in layer thicknesses.
In order to isolate the effects of layer thickness variation, and to exclude effects such as
nonlinearity and anisotropy, all deflection data were generated by means of a layered elastic
Jooste, Kekwick and Muthen 3
program for which the exact input values (i.e. moduli, Poisson’s ratios and layer thicknesses)
were known.
Layer modulus values were then backcalculated using only the average (design)
thicknesses. By comparing the backcalculated moduli to the values originally used to generate
the deflection bowls, an indication of the error due to layer thickness variation could be obtained.
Three nominal structures were used in this study, described in the following section, and
For each layer of each structure one hundred layer thickness variants were
generated, having a mean equal to the nominal design thicknesses and following
a normal distribution. Standard deviations were chosen such that the generated
(ii) Generate deflection bowls for each variant of the nominal structural
For each structure, the generated layer thickness variants and the assumed moduli
shown in Figure 1 were used to generate one hundred deflection bowls. For each
layer the modulus and Poisson’s ratio was kept constant and only the layer
(iii) Derive moduli for each variant of the nominal structure based on design
thicknesses
Backcalculation on these generated deflection data was performed using only the
design thicknesses as input for layer thickness. MODULUS (1) was used for all
backcalculation.
The derived moduli were then compared with those used originally to generate
Jooste, Kekwick and Muthen 4
Figure 1 shows the three nominal pavement structures analysed, together with the
modulus values assigned to each material. The structures were chosen from the South African
pavement design catalogue (7), and represent three different classes of road.
Structure 1 is a relatively heavy structure for high volume major arterial routes,
comprising a high quality crushed stone base on a thick cement-stabilized subbase, with hot-mix
Structures 2 and 3 are designed for successively lower utility roads, respectively. The
former would be typical of a secondary feeder route, in this case designed for a traffic loading
of 0.3 to 1 million ESALs, with the latter typifying a very low volume local road (design
trafficking 0.003 to 0.010 million ESALs). While both use a double surface treatment, they
differ in the base and subbase layer thicknesses, the quality of the granular materials and the use
Typical layer modulus values were chosen based on the ranges recommended in the South
African Mechanistic Design Method (8) for the various materials. In the case of the thin hot-mix
asphalt surfacing for Structure 1, a modulus value of 2,500kPa was assumed. Poisson’s ratio
values of all materials were kept constant at 0.40, which would not necessarily represent expected
actual values but was considered expedient for this study, and unlikely to have any significant
It should be noted that the structures shown in Figure 1 are normally constructed on well-
Jooste, Kekwick and Muthen 5
prepared selected and upper subgrade layers. For the purpose of this analysis, however, these
layers were not modelled individually but were combined with the subgrade to form one single
support layer. The subgrade modulus used for Structure 1 was 100 MPa, while that used for
Structures 2 and 3 was 80 MPa, in each case the values regarded as providing realistic
representations of actual conditions. This simplification still gives a good representation of the
actual support conditions, allowing faster processing without jeopardising the validity of the
comparison.
The thicknesses shown in Figure 1 are the nominal design thicknesses. In the generation
of deflection data these layer thicknesses were varied while the selected moduli values remained
constant, and for each structure deflection bowls were generated using the layer thickness
variants. The simulated FWD load was kept constant at 40 kN, which corresponds to a plate
pressure of 565 kPa over the 300 mm diameter circular load footprint.
In South Africa, the allowable variation of layer thicknesses for road construction are
given in the Standard Specifications for Roads and Bridge Works (9). Limits are placed on
variations from the specified nominal thickness in terms of average values, 90% of sampled
values and for individual localised points. Table 1 gives these limits, designated respectively
Davg,, D90, and Dmax. For the prescribed sampling regime, requiring at least 30 measurements of
a) at least 90 per cent of all thickness measurements taken are equal to or thicker than the
b) the mean layer thickness of the sample lot is not less than the specified layer thickness
Localized areas where the sampled thickness is less than the specified thickness minus
To generate the layer thickness variants it was first assumed that all layer thicknesses will
be normally distributed, with a mean equal to the design thickness. The standard deviation of
the distribution was then selected such that a random sample of 50 data points met the specified
tolerance levels.
Table 2 summarizes the sampled thickness parameters for each pavement structure,
indicating their compliance with the specification requirements, and Figure 2 shows an example
of the distribution of the generated layer thicknesses (for the 150mm base of Structure 1, in this
case). Figure 3 illustrates the entire variation in layer depths for Structure 1 for all 100 thickness
Table 2 shows that all layers essentially conform to the specified tolerances for layer
thicknesses, except one sampled point was outside specification on the subbase of Structure 1.
Although such localized positions would in practice have to be repaired, the layer thicknesses
were not changed for the purpose of this analysis, as the other thickness parameters were well
within specifications.
It should also be noted that the selected standard deviations, necessary to ensure that the
thickness variations comply with the specification, were considered to be low. For example,
typical standard deviations on actual constructed layer thicknesses reported by Darter et al (10)
were all greater than the standard deviations that had to be used to conform to the South African
standard specifications (9). Thus a pavement with standard deviations similar to those reported
by Darter et al (10) would not have conformed to the South African standard specifications. It
is possible, however, that the assumption of a normal distribution for layer thicknesses is not
The MODULUS backcalculation program has the capability to determine the depth to
stiff layer, or subgrade thickness (11). If this depth is used in the backcalculation model then
subgrade moduli estimates are normally more realistic than for models that assume a semi-
infinite subgrade. In the subsequent backcalculation of data, the correct subgrade thickness was
used as input to MODULUS, instead of using the calculated depth to stiff layer supplied by the
program. This was decided since the objective of this study is not to evaluate the efficiency of
the depth-to-stiff layer calculation procedure in MODULUS, but rather to determine the effect
To broaden the scope of the study, two cases for which different depth to stiff layer
conditions apply were analysed. In the first, a set of deflection data was generated by keeping
the thickness of the subgrade fixed at 1500 mm. In the second case the subgrade modulus was
varied according to a normal distribution with a mean of 1500 mm and with a standard deviation
of 150 mm (i.e. a coefficient of variation of 10 per cent was assumed). This second scenario is
considered to lead to deflection bowls that are more similar to those that would be measured in
the field, where the depth to stiff layer varies from point to point.
ANALYSIS OF RESULTS
The influence of the generated layer thickness variation was evaluated by comparing the
backcalculated modulus with the modulus that was originally used to generate the deflection data
for each layer and variant. Thus, for each of the 100 data points, for each layer and each
MODgenMODback . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (Equation 1)
Err (%) [ ]×100
MODgen
Jooste, Kekwick and Muthen 8
It should be noted that there is also an error in the backcalculation process that is not
related to variations in layer thickness, but is due only to errors in matching generated
(representing ‘measured’) and calculated deflection bowls. This error was determined for each
structure by generating one deflection bowl, using the design layer thicknesses, and then
backcalculating the moduli from that deflection bowl. A comparison of the actual and
backcalculated moduli for that deflection bowl therefore gives the error due only to inefficiencies
in the particular backcalculation procedure used. Table 3 gives the backcalculation errors for
each structure, together with the moduli ranges used in all backcalculations, based on such
comparison.
It is clear from Table 3 that the moduli values backcalculated by MODULUS generally
compare very well with those used to generate the deflection data. Except for one case, the
backcalculated moduli are in all cases within 2 percent of the actual moduli. Since the version
of MODULUS used for this study utilized imperial units, deflections and backcalculated moduli
had to be converted to metric units, and it is also therefore possible that a small roundoff error
is contained in the backcalculated moduli. Nonetheless, the results of this comparison clearly
show that, for the pavements analysed here, any errors larger than 4 per cent are likely to be
DISCUSSION OF RESULTS
General
The backcalculation errors due to variations in layer thicknesses are summarized in Table
4 which shows the frequency of occurrence (as a percentage) of error ranges calculated from
Equation 1. It should be noted that the Table 4 shows the effects of both assuming a fixed
subgrade depth and of a varying subgrade depth. In general, the errors due to layer thickness
In all cases the subbase proved to be the most sensitive to small variations in layer
thicknesses. This was especially so in the case of the cemented subbase of structure 2. For this
layer and structure, more than half of the backcalculated moduli had errors in excess of 20 per
cent. The use of a varying subgrade depth in genereting deflection data always increased the
Table 5 summarizes the average backcalculation error due to variations in layer thickness.
Since most pavements will exhibit a significant variation in the apparent depth to stiff layer, the
case where subgrade depth was varied is probably the most representative of real pavement
conditions. It can be seen from Table 5 that, for the case where subgrade depth was varied, the
average error for all layers and all structures was always in excess of 5 per cent. In the case of the
Table 6 summarizes the average backcalculated modulus of each layer for each pavement
structure. It is interesting to note that, despite the large variation in backcalculated moduli at
individual data points, the average backcalculated moduli for the data set compares very well
Structure 1
For Structure 1, the heaviest structure with a high quality crushed stone base on a thick
cement-stabilized subbase, the largest modulus errors arising from layer thickness variation
occurred for the stabilized subbase. For this layer, more than 45 per cent of all data points had
errors larger than 10 per cent. In contrast, the base layer seemed relatively insensitive to layer
thickness variations, with less than 10 per cent of the data having errors greater than 10 per cent.
The subgrade modulus also seemed relatively insensitive to small variations in the
pavement layer thickness, with all errors being below 10 per cent for the case where the subgrade
thickness was kept constant. In the case where subgrade stiffness was varied, significantly larger
errors occurred, with more than 20 percent of all data points having errors that are in excess of
20 per cent.
Structure 2
Except for the subgrade, the backcalculation errors due to variations in layer thickness
were generally the highest in the case of Structure 2. As with Structure 1, the subbase modulus
proved to be the most sensitive to small variations in layer thickness. For this layer, more than
50 per cent of all backcalculated moduli had errors in excess of 20 per cent. The base modulus
also contained significant errors, with more than 40 per cent of all data having errors larger than
10 per cent. Figure 5 shows the variation of the backcalculated moduli of the subbase of
structure 2 about the correct modulus value (i.e the modulus value that was used to generate
deflections).
For this structure, the subgrade proved to be insensitive to pavement layer thickness
variations as long as the subgrade thickness was kept constant. When the subgrade thickness was
varied, approximately 20 per cent of all data had errors in excess of 10 per cent.
Jooste, Kekwick and Muthen 11
Structure 3
It is perhaps surprising that Structure 3, which is the thinnest pavement structure that was
analysed, generally proved to be less sensitive to layer thickness variations than the other two
structures. For Structure 3, the subbase also exhibited the largest errors in backcalculated moduli.
For the base, no data points had errors larger than 10 per cent when the subgrade thickness was
kept constant. When subgrade thickness was varied, nearly 30 per cent of all data points had
that are acceptable in terms of construction tolerance specifications, was investigated. The study
revealed that even small variations in layer thicknesses can significantly affect the values derived
` The backcalculation error due to small variations in layer thicknesses is at least 5 per cent
` For the structures investigated in this study, the average backcalculation error for the
` Not all layers are equally sensitive to layer thickness variations. For each of the
structures evaluated the subbase appeared to be the most sensitive to small variations in
` The results of this study demonstrate that thin, stiff layers such as stabilized subbases are
very sensitive to small variations in layer thicknesses, as might be expected. For one of
the structures studied more than 60 per cent of all data points had backcalculation errors
` The backcalculation error due to small variations in layer thickness was generally lowest
on the low standard pavement structure (Structure 3), even though this structure had fairly
thin layers. A probable reason is that this pavement did not have an inverted structure
with a stiff stabilised subbase, and the moduli of all layers were generally quite similar
` Despite the relatively large variations in backcalculated moduli at individual data points,
the average backcalculated moduli for the data set compared well with the correct
modulus value (i.e the modulus value used to generate the deflection bowls).
The above conclusions highlight that, even when the average layer thickness is
determined correctly for a properly constructed pavement, backcalculated moduli may contain
significant errors due to small and probably unavoidable variations in layer thickness. Clearly
an incorrect estimate of layer thickness, which may be attributable to numerous practical factors
such as insufficient data, incorrect data or poor sampling, will result in even larger errors in
backcalculated moduli.
This study also suggests that there are limits to the precision that can be attained in the
backcalculation of moduli. Thus the effort involved in trying to get a better fit between measured
and calculated deflections may be spent more effectively by trying to get a better indication of
layer thicknesses.
Jooste, Kekwick and Muthen 13
REFERENCES
1. Modulus Reference
2. Evercalc Reference
3. Elmod Reference
6. Weslea Reference
7. TRH4
9. CSRA Specification
LIST OF FIGURES
LIST OF TABLES
5. Average Moduli Backcalculated for the Entire Data Ser for each Pavement Section
Jooste, Kekwick and Muthen 15
Selected Minimum
Layer Standard Sampled
Pavement and nominal Deviation Average Thickness % of Points
Structure thickness (mm) (mm) a (mm)a Failing D90b
1 Asphalt Surface 3.5 40.0 33.3 4
40 mm (38) (32)
Base 16 153.2 124.3 1
150 mm (145) (123)
Subbase 15 248.2 218.6 4
250 mm (245) (223)
2 Base 12 127.2 102.0 2
125 mm (120) (98)
Subbase 14.5 147.8 126.3 5
150 mm (145) (123)
3 Base 12 125.3 104.5 0
125 mm (120) (98)
Subbase 12 126.1 103.4 2
125 mm (120) (98)
a
Values in brackets denote allowable minima according to Table 1.
b
Denotes the percentage of sampled data points less than design thickness minus D90 (specified
in Table 1). A maximum of 10 per cent is allowed.
Jooste, Kekwick and Muthen 16
Table 6 Average Moduli Backcalculated for Entire Data Set for Each