A RELIABILITY-BASED MIX DESIGN AND ANALYSIS

SYSTEM FOR MITIGATING FATIGUE ISTRESS

John T. Harvey John A. Deacon
Institute of Transportation Studies Department of Civil Engineering
University of California, Berkeley University of Kentucky
Richmond Field Station Lexington, KY 40506
1353 S. 46th Street, Bldg. 480
Richmond, CA 94804-4603

Akhtar A. Taybali Rita B. Leahy

Department of Civil Engineering Department of Civil Engineering
North Carolina State University Oregon State University
Raleigh, NC 27695-7908 Corvallis, OR 9733l-2302

Carl L. Monismith
Department of Civil Engineering
University of California, Berkeley
Berkeley, CA 94720

Abstract. A reliability-based mix design and analysis depends on critical interactions between mix properties
system has been developed for mitigating fatigue distress and in-situ conditions, thus providing not only sensitivity
in asphalt pavements. Design decisions consider not only to mix behavior but also sensitivity to construction
fundamental mix properties but also the level of design performance, in-situ traffic, climate, and the pavement
traffic, the temperature environment at the site, the structure. The mix design and analysis system and its
pavement structural section, laboratory testing and initial calibration for California conditions have been
construction variabilities, and the acceptable level of risk. described in detail elsewhere (Deacon et al., 1994b). The
This paper describes the mix design and analysis system, current study continued to focus on California experience
discusses its calibration, illustrates its use, and assesses as an example of how the system can be calibrated to
the consistency of current California design practice vis- reflect local conditions. It has enabled improvements to
a-vis the control of fatigue distress. It reveals possible be made which incorporate additional laboratory test
inconsistencies in ‘current California design practice, results, better reflect California pavement design
discusses effects of construction variability, highlights experience and climatic conditions, and consider effects
advantages of rich-bottom pavement structures, and of construction variability. Following a brief overview of
illustrates the integration of mix and structural the design and analysis system, these refinements are
components into a reliability-based mix design process. discussed in depth. The paper concludes by exploring
several of the implications of this work on design and
Keywords. Asphalt, fatigue, mix, design, reliability construction practice.

INTRODUCTION MIX DESIGN AND ANALYSIS SYSTEM

An important development of the recently completed General Description. Mix design typically involves an
Strategic Highway Research Program Project A-003A iterative search for an economical combination of
was a mix design and analysis system which explicitly ingredients which meets performance requirements
accounts for fatigue distress. The system relies on while, at the same time, minimizes the risk of premature
laborator-y flexural-fatigue testing~7 for mix evaluation and failure. Compromise is necessary because the various
incorporates ah analysis system for properly interpreting distress mechanisms-permanent deformation. fatigue.
test results. It recognizes that mix pcrformance in situ thermal cracking, bleeding, ravelling, etc. - often impose
contradictory demands. The following discussion focuses in which N= the number of laboratory load repetitions to
only on fatigue performance. As presented herein, the failure under the anticipated in-situ strain level, ESALs=
fatigue design and analysis system assumes that a trial the number of equivalent, 80 k N (18,000-pound) single
mix has been selected, and it evaluates the likelihood that axle loads expected in the design lane during the design
this mix will satisfactorily resist fatigue cracking in the period, TCF = the temperature conversion factor, M= the
design pavement under anticipated in-situ conditions. reliability multiplier, and SF = the shift factor. When a
In essence, a mix is expected to perform mix under consideration does not meet this requirement,
satisfactorily if the number of load repetitions sustainable the designer has a wide range of options including
in laboratory testing exceeds the number of load adjusting the mix by adding more asphalt and/or
repetitions anticipated in service. To minimize laboratory reducing the air voids, using a different asphalt,
costs, testing is at an accelerated rate and, for normal aggregate, or aggregate gradation, increasing the
asphalts, at a single temperature. Using the University of pavement thickness, and even allowing an increased risk
California, Berkeley (UCB) controlled-strain, flexural- of premature failure.
fatigue apparatus, testing can usually be accomplished For especially important and expensive projects, a
within a 24- to 48-hour period. The strain at which the more comprehensive design process may be warranted.
number of laboratory repetitions must be estimated is This process is required for new or unconventional
computed using multilayer, elastic theory. For this paving materials. As a part of this process, fatigue tests
computation, the strain of interest is the maximum are performed at multiple temperatures [in the range
principal tensile strain at the bottom of the asphalt-bound l0-30°C (50-86°F)]. Rather than using temperature
layer in the design pavement. A 40 kN (9,000-pound), conversion factors as in the procedure described above, a
dual-tire load is applied, and the mix stiffness is taken detailed analysis is performed based on the hourly
from the laboratory test results at the laboratory test temperature distributions expected in situ. With these
temperature. temperatures and the corresponding traffic distributions,
Traffic is represented by the number of equivalent the linear-sum-of-cycIe-ratios cumulative damage
single axle loads (ESALs) in the critical or design lane hypothesis is used to evaluate the adequacy of the
during the design period. Because these ESALs pavement structure containing the non-conventional mix.
accumulate within a-mixed temperature environment, it is Procedures are available for determining
necessary to apply a factor, herein termed the temperature conversion factors, reliability multipliers,
temperature conversion factor, to convert design ESALs and shift factors, and preliminary estimates have been
to their equivalent at a single temperature, that used in developed (Deacon et al., 1994a and 1994b). Included in
the laboratory testing. Experience has shown that it is this paper are additional refinements and extensions.
also necessary to apply a shift factor, accounting for a
host of factors such as traffic wander, crack propagation Laboratory Testing. Fatigue performance and stiffness
time, frequency of loading, etc., to assure that load of the mix under evaluation are measured by repeated
repetitions in the field are commensurate with those in flexure of beam specimens. The test apparatus,
the laboratory. developed as part of SHRP A-003A (Tayebali et al.,
Because of uncertainty or variability in all of the 1994a and 1994b), subjects these specimens to third-
measurements, simulations, and predictions, there is point, controlled-strain flexure. A 10 Hz haversine
some risk that a mix will fail in service even if its deformation wave is continuously applied to the beam
laboratory resistance is determined, by the aforedescribed until failure, defined as the point at which beam stiffness
process, to exceed the design loading. Fortunately the has been reduced to one half its initial value. Testing
risk of failure can be limited to a tolerable level by usually involves two or three replicate tests at each of
applying a reliability multiplier to the design loading two strain levels. Upon completion of testing, a least-
before comparisons are made with the laboratory squares regression is performed to find the coefficients a
resistance. The fatigue design and analysis system and b in the fatigue life (N) versus tensile strain (E)
incorporates risk assessments in this way. equation:
In summary a mix is deemed to be suitable for use in
N”OEb (2)
the selected pavement structure to mitigate fatigue
cracking when:
Laboratory preparation of test specimens includes
subjecting the loose mix to four hours of short-term oven
N > ESALs*TCFd4 aging at 135°C in a forced-draft oven. This short-term
- (1)
SF oven aging was recommended by SI-IRP A-003A
researchers for duplication of the combined aging effects revised slightly to better accommodate input/output
of construction and up to two years of aging thereafter requirements and to replace the method for computing
(Bell et al., 1994). Rolling wheel compaction is used in extraterrestrial radiation with one thought to be more
order to produce specimens with performance suitable.
characteristics similar to those of field-compacted mixes. Pavement temperature simulations spanned 1 O-year
Test beams, 380 mm in length, 64 mm in width, and periods for ‘Santa Barbara and Daggett Airports and a
5 1 mm in height (15 in., 2.5 in., and 2.O in., respectively) 6-year period for Blue Canyon. Table 1 presents relevant
are cut from the slabs thus formed. location information and climatic summaries for the three
sites. Important parameters used in the temperature
TEMPERATURE CONVERSION FACTOR simulations are summarized in Table 2. The simulations
produced detailed, long-term temperature profiles for
The temperature conversion factor (TCF) is a each of the two pavements. Because the laboratory
multiplicative factor which converts the number of testing of this study was conducted at 19°C, it was
design load repetitions occurring in the mixed necessary to compute TCFs for this temperature. Also
temperature environment in situ to its equivalent at a determined were the critical temperature, that
single temperature. The ability to make such a temperature at which most fatigue distress occurs, and
conversion is critically important for routine work, TCFs to the critical temperature. Results of these
because it reduces the requisite laboratory testing and computations, summarized in Table 3, confirm prior
structural analysis to a single temperature. The TCF has findings that TCFs are both site- as well as pavement
been shown to be dependent on both the pavement structure-specific.
structure and the thermal environment (Deacon et al., They also confirm prior findings that critical
1994a). At the same time, it has been hypothesized that temperatures are larger for 203 mm (8-inch) pavements
the TCF may be relatively unaffected by mixture than for 102 mm (4-inch) pavements and, for both
characteristics, particularly for asphalts of normal thicknesses, are considerably in excess of the test
temperature sensitivity. It remains for future study to temperature of 19°C. Testing at critical-temperature
determine the limits within which this hypothesis is valid. levels would be advantageous because possible errors
Deacon et al. (1994a) have developed a procedure due to temperature-sensitivity abnormalities and due to
for calculating TCFs and have applied it to two necessary extrapolations are minimized. Nevertheless the
hypothetical pavement structures located within nine convenience of testing at cooler temperatures and
geographic regions of the United States. The desire to uncertainties about the legitimacy of fatigue
more specifically address California climates together measurements at higher temperatures in the critical-
with advancements in the computation of pavement temperature range have dictated the choice at UCB of
temperature profiles led to the additional computations testing temperatures at or near 20°C.
reported herein. Because TCFs have been computed for only two
Three California locations, representative of a surface thicknesses, the precise influence of pavement
variety of climatic conditions, were examined including: structure on TCFs is not now known. To avoid
Blue Canyon in Placer County (mountain environment); excessively large TCFs that would result from linear
Daggett Airport in San Bemadino County (desert extrapolation for thick pavements, the logarithmic
environment); and. Santa Barbara Airport in Santa relationships depicted in Figure 1 are recommended. The
Barbara County (coastal environment). At each location, TCFs can be computed by the following:
two hypothetical pavements were examined--the same
two used earlier and described in detail by Deacon et al. TCF=aIn (d)+ b (3)
(1994a)--having 102 mm (4.0 in.) and 203 mm (8.0 in.) in which d = asphalt concrete thickness in centimeters
surface courses. and the regression constants, a and b, are as follows:
The asphalt mix was thought to be representative of
those used in California, and laboratory stiffnesses and
fatigue lives were measured and characterized during the Location a b
recent SHRP A-003A investigation. Pavement
Santa Barbara 1.754 -2.891
temperature profiles were simulated using the climatic-
Daggett 2.102 -3.884
materials-structural (CMS) pavement analysis model
Blue Canyon 1.448 -2.475
originally developed at the University of Illinois
(Dempsey et al., 1985, and Herlache et al., 1985) and
Table 1. Location and climatic summary of California sites
Attribute Site
~~
Santa Barbara Daggett Blue Canyon
Latitude (°) 34.433 34.867 39.283
Elevation (m) 1 179 490
Constant deep ground temp. (°C) 16.7 21.1 13.9
Period 1984-1993 1983,1985-1993 1983-1988
Average minimum daily air 93
. 11.8 66
.
temperature (“C)
Average maximum daily air 21.5 27.5 14.9
temperature (°C)
Average daily sunshine (%) 67.9 81.3 65.3
Average daily wind speed (km/h) 9.6 18.0 6.9

Table 2. Selected temperature simulation parameters’

Layer Property Calibrated Value
Asphalt mixture Thermal conductivity 0.8 BTU/hr-ft-F (unfrozen, frozen, freezing)
Heat capacity 0.22 BTU/lb-F (unfrozen and freezing)
1.2 BTU/lb-F (freezing)

Unit weight 145 pcf

Water content 2%

Subgrade Dry density 120 pcf

Dry heat capacity 0.29 BTU/lb-F

Other Maximum allowable convection 3 BTU/hr-ft2-F

coefficient

Emissivity factor 0.85

Absorptivity factor 0.90

Geiger radiation factor A 0.77

Geiger radiation factor B 0.28
Vapor pressure 5 mm mercury
Cloud base factor for back radiation 0.85
“Only English units are shown because the computer program and data files are expressed in English units.
Table 3. Temperature conversion factors and critical temperatures
Thickness of Parameter Site
asphalt layer
Santa Barbara Daggett Blue Canyon
102mm TCF (to 19°C) 1.176 0.989 0.883
(4 inches)
Percent of damage in 5°C 16.6 12.8 13.8
range about 19°C
Critical temperature (°C) 27 30 28
TCF (to critical temperature) 0.655 0.587 0.481
Percent of damage in 5°C 34.6 24.4 30.2
range about critical
temperature
203 mm TCF (to 19°C) 2.392 2.446 1.887
(8 inches)
Percent of damage in 5°C 8.4 5.8 7.3
range about 19°C
Critical temperature (°C) 29 34 30

TCF (to critical temperature) 0.582 0.546 0.417

Percent of damage in 5°C 49.3 32.9 41.4
range about critical
temperature

4.0

3.5
8
z
E 3.0
E
‘Lz 2.5
$ -Blue Canyon
g 2.0 -mwm*_ Daggett
0 - - - Santa Barbara
5 1.5
E
x 1.0
E
If! 0.5

0.0
4 6 8 10 12 14 16
Thickness of Asphalt Concrete (in)

Figure 1. Effect of Location and surface thickness on temperature conversion factor

RELIABILITY AND VARIABILITY variability reflects a combination of factors including the
inherent variability in fatigue measurements (associated
Decisions about anticipated mix performance cannot both with specimen preparation as well as testing
be made with absolute certainty. Although large safety equipment and procedures), the nature of the laboratory
factors can reduce the likelihood of error, their cost testing program, and the extent of extrapolation
consequences can be considerable. Reliability analysis necessary for estimating fatigue life (using a least-
ensures an acceptable level of risk in mix design without squares, best-fit line) at the design strain [Link](lnN)
the cost of excessively large safety factors. is calculated as follows (Deacon et al., 1994b):
Reliability is considered, herein, to be the probability
that the mix will provide satisfactory performance
var(lnN)=s 2[ l+ $$I+.+.;J (5)
throughout the design period. The reliability level for
each specific mix design is set by the designer. Larger
levels of reliability reduce the chances of accepting in which s2 = the variance in logarithm of fatigue-life
deficient mixes; however, the tradeoff is the potentially measurements, n = number of test specimens, X =
larger cost associated with reducing the number of
ln(in-situ strain) at which ln(N) must be predicted, x =
acceptable materials or mixes or increasing the thickness
average ln(test strain), q = number of replicate specimens
of the asphalt concrete.
at each test strain level, and x p = ln(strain) at the pth test
Reliability is introduced in the mix design and
strain level. For most of the mixes considered herein, six
analysis system by a reliability multiplier, M, which is
specimens were tested (n = 6) including three (q = 3) at
calculated as follows:
each of two strain levels (xp = 150 and 3 00 microstrain).
Subsequent attention turned to the effects of
M - e ZJvar(ln N>+var(ln ESALS) construction variability on var(lnN) Two components of
(4)
construction variability were considered, those affecting
in which e = the base of natural or Naperian logarithms, Z mix behavior and those affecting pavement structure
= a factor depending solely on the design reliability, behavior.
var(lnN) = the variance of the logarithm of the laboratory Potentially important mix variables include asphalt
fatigue life estimated at the in-situ strain level under the content and air-void content. Fortunately, the necessary
s t a n d a r d 4 0 kN (9,000-pound) w h e e l l o a d , a n d data for quantitative evaluation of these variables is
var(lnESALs) = the variance of the estimate of the available, namely, construction variances and, for a
logarithm of the design ESALs. Z is related to design typical mix, relationships with mix stiffness and fatigue
reliability as follows: life (Harvey et al., 1996). Other potentially important
Design reliability (percent) Z mix variables, such as aggregate gradation, were
excluded from consideration because construction
95 1.64 variances and/or stiffness and fatigue life relationships
90 1.28 were unknown.
80 0.84 Potentially important structure variables include
60 0.253 thickness of the asphalt concrete layer and stiffnesses and
50 0.000 thicknesses of other supporting layers. Obviously, asphalt
concrete thickness was a critical factor to be considered,
and construction variance data was fortunately available.
The variability associated with forecasts of design
The supporting layers are more problematic because
ESALs is unknown. In the absence of better information,
knowledge of the variabilities of foundation layer
var(lnESALs) has been assumed herein to be 0.300 1 .
thicknesses and moduli seems to be limited, and
First efforts to compute var(lnN) focused on
complexity of the necessary analysis increases with
variability associated with the testing program. Testing
increases in the number of layers. As a result, it was
decided to simplify the process by identifying an
1 equivalent single layer whose effects were similar to the
This estimate is based primarily on the authors
multilayer supporting layers. The equivalent single layer
judgment. As a point of reference, the AASHTO design is characterized by its modulus, and the variance of this
guide suggests that actual traffic may be 1.6 times that modulus was evaluated from
equivalent
predicted at a one-standard-deviation level. Assuming a
post-construction falling weight deflectometer (FWD)
log normal distribution, this corresponds to a
measurements.
var(lnESALs) of about 0.22.
As currently developed var(lnN) is thus the sum of in which ?VS5’~ = within sum of squares for factor level
three components as follows: combination k (air-void content, asphalt content, and
strain level), n = number of observations (96), d =
var(lnN) =A!?*2 +s,+s,
2 2 (6) number of factor level combinations (30), InN, =
logarithm of measured fatigue life for specimen i, mk =
in which s12 = the testing variance, s,~ = the mix variance, average logarithm of measured fatigue lives of specimens
and s,’ = the structure variance. While direct for factor level k , and nk = number of replicates at factor
computations yield s12 (Equation 5), computing s,’ and level k .
ss2 is more difficult. Fortunately, Monte Carlo simulation The best estimate of the overall sample variance of
is a convenient and relatively quick way to obtain the the natural logarithm of the fatigue life was calculated
necessary estimates. For the design pavement structure using the aforedescribed procedure to be 0.220. The
and test temperature, each simulation proceeds as sample variance obtained in the SHRP A-003A expanded
follows. A random selection is made of asphalt content tested program, using the same testing equipment and
and air-void content. The mix stiffness is determined procedures, was 0.152 (Tayebali et al., 1994a). Because
from these random selections, and a “‘mean” pavement of greater replication in the current study (three versus
strain is calculated (ELSYM5). A randomly selected two in A-003A) and smaller strain levels (150 and 300
adjustment is made to this strain level to account for microstrain versus 400 and 700 microstrain in A-003A),
random variations in surface thickness and support 0.220 seems to be the best possible estimate of s2 at the
modulus. The fatigue life is then computed from the current time. It was used in developing the shift factors in
random selections of tensile stain and the asphalt and air- the mix design and analysis model and is recommended
void contents. The process is repeated until the desired for use for mix design purposes.
degree of convergerice has been reached. In work to date, Estimates of the variances of asphalt content, air-
distributions of all of the four construction variables have void content, asphalt-concrete thickness, and foundation
been assumed to be normal. The Monte Carlo simulations modulus were obtained from a combination of literature
can and have been extended to include testing variability review, moduli backcalculations of FWD measurements,
as well as construction variability. and unpublished data recently collected as a part of the
The individual variances necessary for the WesTrack project. Summary results are presented in
simulations include the testing variance (s:) and the Table 4. These totals include not only materials and
variances of asphalt content, air-void content, asphalt- construction components but also components resulting
concrete thickness, and foundation modulus. from testing and sampling. The latter components must
The testing variance was determined through a be removed from the variance estimates in order to
recently completed laboratory test program (Harvey et isolate materials and construction effects: Table 5
al., 1996). The replication which was built into the summarizes the necessary data.
experiment design permitted computation of the inherent Finally, Table 6 summarizes the quantities used
variance (s2) in the logarithm of fatigue-life herein to represent reasonable estimates of
measurements. The experiment design involved 1 5 materials/construction variability associated with
different mixes (five asphalt contents and three air-void conventional construction practice. The equations for
contents) tested at each of two strain levels with estimating the standard deviation of asphalt-concrete
nominally three replicates at each of the 30 combinations thickness were developed herein as an approximate way
(a few small-strain tests had four replicates). A total of 96 to handle multilift construction. Among the assumptions
observations were included in calculating the sample made in their development was that the coefficient of
variance using the following relationships: variation of thickness in single-lift construction is about
14 percent.
2 wss k
Using these variance estimates, Monte Carlo
simulation was used to produce estimates of var(ln N ) for
S2 = k =l
n - d
(7)
and
2WesTrack refers to a Federal Highway Administration

WSSk = 2 (hNi - I II Nk)' (8)

Project, “Accelerated Field Test of Performance-
Related Specification for Hot-Mix Asphalts
i=Z Construction,” which incorporates an experimental
road test facility recently constructed in Nevada.
Table 4. Construction variation of mix and structural characteristics
Property Measure of variation Value or range Source
Asphalt Standard deviation 0.15-0.44% Table 12.46 (Epps, 1996) _
content
Standard deviation 0.190.4% Individual WesTrack sections
Standard deviation 0.3 1% WesTrack composite
Standard deviation 0.3% Table 3 (Benson, 1995)
--
Air-void Standard deviation - 0.9-1.9% Table 12.55 (Epps, 1996)
content
Standard deviation 0.4-l .5% Individual WesTrack sections

Standard deviation 1.5% WesTrack composite

Standard deviation I .94% Table 3 (Benson, 1995)
Thickness Coefficient of variation 12.5-15% Table 12.58 (Epps, 1996)
Standard deviation O-O.5 cm Individual WesTrack sections
Standard deviation 0.58 cm WesTrack composite
Standard deviation 0.99 cm Table 3 (Benson, 1995)
Foundation Coefficient of variation 11.3 -14.7% H V S test sections at UCB
modulus
Coefficient of variation 17344.7% Segment of AA highway in KY

Coefficient of variation 3.6-17.7% Individual WesTrack sections

Coefficient of variation 14.2-28.5% WesTrack composite

Table 5. Materials/construction component of total construction variance

Property Materials/construction component Source
(0
OO

Asphalt content 40 Figure 7 (Granley, 1969)

61 Table 3 (Benson, 1995)

Air-void content 60 Table 8 (inferred) (Granley, 1969)

90 Table 3 (Benson, 1995)

Thickness 95 Table 3 (Benson, 1995)

Foundation modulus 70 Assumed

Table 6. Variation of mix and structural characteristics for Monte Carlo simulations
Property Total standard deviation Percentage of variance due Materials/construction
to materials/construction component of standard deviation
Asphalt content 0.30% 40 0.19%
Air-void content 1.6% 60 1.2%
Surface thickness 0.200 * To-69 cm 80 0.173 0 Toa@ cm
Foundation modulus 30% (coef. of variation) 70 25%
2
1.8
1.6
1.4
El Structure
1.2
1
0.8 0 Mix
0.6
0.4
0.2 m Test
0

Pavement Section

Figure 2. Major components of testing and construction variance

18 hypothetical pavements designed according to indices. The structures component of construction

Caltrans procedures and representing three traffic levels variability is slightly greater for thinner than for thicker
(TI of 7, 11, and 15), two base types (cement treated base pavements while a reverse trend is observed for the mix
[ctb] and aggregate base [ab]), and three subgrade component.
strengths (R-value of 5, 20, and 40) (Table 7). The ctb
contains a maximum 2.5 percent cement by weight of SHIFT FACTOR
aggregate, insufficient to strongly bond aggregate
particles. The Valley asphalt- Watsonville granite mix had As a result of such factors as traffic wander, crack
target asphalt and air-void contents of 5 and 8 percent, propagation time, frequency of loading, etc., highway
respectively; stiffness and fatigue relationships have pavements have been found to sustain from less than 10
been previously reported (Harvey et al., 1996, and to perhaps as many as 100 times the number of load
Harvey and Tsai, 1996). applications that are estimated by procedures similar to
Figure 2 illustrates the level and components of those used herein before they become seriously
var(lnN) for each of the 18 structures. The testing portion distressed. As a result, laboratory estimates of fatigue life
of the total variance increases significantly with the can be compared with service estimates of ESALs only
increases in pavement thickness associated with larger after applying a suitable shift factor. Using AASHTO
traffic indices principally because of the increased level design guidelines as a basis. SHRP A-003A studies led to
of extrapolation to the smaller strain levels. Construction the recommendation of shift factors ranging from 10 to
variability is important for all of the pavements but is a 14 depending on the amount of surface cracking
much greater portion of the total variability for the less considered to be tolerable (Deacon et al., 1994b). A shift
substantial pavements which accompany smaller traffic factor of 10, for example, means that each application of
Table 7. Characteristics of hypothetical pavement structures
Traffic Subgrade Layer Class B cement-treated base Class 2 aggregate base
index R-value
Thickness Stiffness Poisson’s Thickness Stiffness Poisson’s
ratio ratio
mm in MPa psi in MPa psi
7 5 Surface 91.4 3.6 Varies Varies 0.40 91.4 3.6 Varies Varies 0.40 *
Base 168 6.6 276 40,000 0.30 168 7.2 207 30,000 0.45
Subbase 259 10.2 138 20,000 0.45 259 10.2 138 20,000 0.45
Subgrade 26.5 3,850 0.50 26.5 3,850 0.50
20 Surface 91.4 3.6 Varies Varies 0.40 91.4 3.6 Varies Varies 0.40
Base 168 6.6 276 40,000 0.30 183 7.2 207 30,000 0.45
Subbase 152 6.0 138 20,000 0.45 I 152 6.0 138 20,000 0.45
Subgrade 84.1 12,200 0.50 84.1 12,200 0.50
40 Surface 91.4 3.6 Varies Varies 0.40 107 4.2 Varies Varies 0.40
Base 183 7.2 276 40,000 0.30 168 6.6 207 30,000 0.45
Subbase I
Subgrade 161 23,400 0.50 161 23,400 0.50
11 5 Surface 198 7.8 Varies Varies 0.40 259 10.2 Varies Varies 0.40
Base 198 7.8 220 32,000 0.30 152 6.0 172 25,000 0.45
Subbase 427 16.8 138 20,000 0.45 335 13.2 138 20,000 0.45
Subgrade 26.5 3,850 0.50 26.5 3,850 0.50
20 Surface 213 8.4 Varies Varies 0.40 244 9.6 Varies Varies 0.40
Base 168 6.6 220 32,000 0.30 152 6.0 172 25,000 0.45
Subbase 259 10.2 138 20,000 0.45 213 8.4 138 20,000 0.45
Subgrade 84.1 12,200 0.50 84.1 12,200 0.50
40 Surface 244 9.6 Varies Varies 0.40 244 9.6 Varies Varies 0.40
Base 152 6.0 220 32,000 0.30 152 6.0 172 25,000 0.45
Subbase
Subgrade 161 23,400 0.50 161 23,400 0.50
15 5 Surface 351 13.8 Varies Varies 0.40 411 16.2 Varies Varies 0.40
Base 152 6.0 172 25,000 0.30 152 6.0 138 20,000 0.45
Subbase 549 21.6 138 20,000 0.45 I 442 17.4 138 20,000 0.45
Subgrade 26.5 3,850 0.50 26.5 3,850 0.50
20 Surface 274 10.8 Varies Varies 0.40 320 12.6 Varies Varies 0.40
Base 259 10.2 220 32,000 0.30 183 7.2 138 20,000 0.45
Subbase 381 15.0 138 20,000 0.45 I 381 15.0 138 20,000 0.45
Subgradc 84.1 12,200 0.50 84.1 12,200 0.50
40 Surface 366 14.4 Varies Varies 0.40 335 13.2 Varies Varies 0.40
Base 152 6.0 172 25,000 0.30 152 6.0 138 20,000 0.45
Subbase 107 4.2 138 20,000 0.45
Subgrade 161 23,400 0.50 161 23,400 0.50
strain in the laboratory is equivalent to 10 applications of
the same strain in situ. ESALs l TCF l M
The most accurate way to develop shift factors is
SF - (10)
N
probably by observing the fatigue performance of full-
scale pavements in test tracks or in accelerated pavement Shift factors were thus determined for each of the 18
loading experiments. Even without such, experimental hypothetical pavements. In the process, ESALs were
data, however, first-order approximations of shift factors estimated from the traffic index by the following
can be developed based on the pavement performance approximate relationship (Harvey et al., 1996):
experience that is captured in specific design procedures.
ESALs = 1.289 5 l low 2 TI 8o2g1g (11)
In the current project, results from laboratory fatigue and
factors suitable for use in mix-design-and-analysis TCFs were computed based on depth of the asphalt
applications. concrete layer, and separate computations were
The mix containing 5-percent asphalt and 8-percent performed for three locations, Santa Barbara, Daggett,
air voids was selected as typical of mixes for which the and Blue Canyon. To compute the reliability multiplier
California thickness design procedure is most applicable. (M), a 90-percent reliability was assumed, and var(ln N )
The performance of this mixture in the 18 hypothetical was based on an s 2 of 0.220, laboratory testing involving
pavement structures (Table 7) was simulated. Such three specimens tested at each of two strain levels (150
simulations produce estimates of the laboratory fatigue and 300 microstrain), and the construction variabilities of
life (N), and the traffic index determines the traffic Table 6. The “laboratory” fatigue life (N) was computed
loading (ESALs). The three traffic levels were 89,800 to from in-situ strain simulations under the 40 kN (9,000-
164,000, 4.5 to 6.6 million, and 64.3 to 84.7 million pound) dual-tire loading.
ESALs, corresponding to traffic indices of 7, 11, and 15, Results of the shift-factor computations show a
respectively. The two base types were Class 2 aggregate dependence of shift factor on both strain level and traffic
base (ab) and Class B cement-treated base (ctb). index (Figure 3). The influence of site location seems to
Subgrade R-values were 5, 20 and 40. The mix design be relatively minor. Unfortunately the nature of the
and analysis system defines the relationship between N relationships of Figure 3 cannot be confirmed from prior
and ESALs for an adequate mixture to be: work since no prior work has been identified which has
investigated variables of possible influence on shift
ESALs l TCF l A4 factors. The real difficulty in this analysis, though, is the
N.- (9)
SF fact that each pavement represented in Figure 3 does not
resist fatigue distress equally well. California design
and, as a result, the shift factor (SF) is given by: relationships do not explicitly treat fatigue distress, and,
while it is possible that most designs are at least
marginally adequate, some almost certainly are

100
90 l Santa
80 Barbara
70
60 w Daggett
50
40
30 A Blue Canyon
20
10
0
0.00000 0.00005 0.00010 0.00015 0.00020 0.00025
Strain

Figure 3. Effect of pavement strain and traffic index on shift factor (90-percent reliability)
0.000050 0‘000100 0.000150
Simulated Pavement Strain
l 50th percentile - Expon. (50th percentile)
2 . ,%*- ;“Power (50th percentile) -Log. (50th percentile)
Figure 4. Effect of pavement strain on shift factor at 50th-percentile levels

overconservative. Design to accommodate the median the nature and thickness of the structural section, the rate
condition should lead to a relatively conservative, first- of accumulation of traffic loading, mode of loading, and
order approximation of shift factors suitable for initial perhaps mix properties as well.
use.
It seems logical to relate the design shift factor to the CONSISTENCY OF CALIFORNIA STRUCTURAL
level of pavement strain. The rate of crack propagation, a DESIGN PRACTICE
principal difference between in-situ and strains
laboratory behavior, is affected by strain level: small the As previously indicated, 18 hypothetical pavements,
not only increase the number of load repetition to crack located in three regions of California and designed using
initiation but also slow the rate of crack propagation as Caltrans procedures, formed the basis for the shift-factor
well. The opposite is expected for large strains. calibrations reported herein. The calibration process
Figure 4 shows the relationship between the 5Oth- underscored differences in expected fatigue behavior
percentile shift factor and the 50th-percentile pavement within a group of pavement sections that had been
microstrain. Each point represents one of the three levels designed to meet the same specific level of traffic
of traffic index. Until such time as more information loading by Caltrans procedures. This finding was not
becomes available, the power relationship is unexpected simply because the Caltrans design
recommended for design purposes. Because of its shape procedure is largely empirical, and no distinction is made
at low strain levels, however, extrapolations should not among the various forms of pavement distress it
be made beyond the limits of available data. The considers. At the same time, the UCB mix design and
recommended design relationship is as follows analysis system provides the opportunity to examine the
consistency of California design practice vis-&-vis the
SF = 3.1833 10 -5 &-1*375g
l (12) prevention of fatigue distress.
for E I 0.000040 For this analysis, the design reliability was set at 90
percent, and stiffness and laboratory fatigue life were
in which E = simulated strain produced by standard wheel calculated for the Valley asphalt-Watsonville granite mix
load at the underside of the asphalt-concrete layer. having 5-percent asphalt and g-percent air voids. UCB
Although the shift factors proposed herein represent ESALs were computed as follows:
an effective point of beginning, future adjustments are
inevitable. Ultimately, shift factors are expected to N4F
ESALs = (13)
depend not only on strain level but also on the extent of TCF l iM
permissible cracking and possibly such added factors as
in which N = the number of laboratory load repetitions to most vulnerable to fatigue cracking appear to include: 1)
failure under the anticipated in situ strain level, SF = the designs for a TI of 15; 2) designs incorporating
shift factor, TCF = the temperature conversion factor, subgrades with R-values of 20; 3) designs for regions
and A4 = the reliability multiplier. California design with climates similar to Santa Barbara; and 4) designs
ESALs were computed using Equation 11. Table 8 incorporating class B cement-treated base. The relatively
summarizes the calculations. An ESAL ratio much less low median ratio for R-value 20 subgrades is of
than one indicates that the UCB fatigue life is particular interest and potential concern. Although the
considerably less than the Caltrans design life and possible significance of these findings is conjectural at
suggests a site-structure combination that appears to be the moment, the analysis demonstrates rather
most vulnerable to fatigue distress in situ. conclusively that, even if all Caltrans designs provide
To identify possible patterns of bias, median ESAL adequate fatigue resistance, they vary considerably in the
ratios were determined for the various levels of each level or extent of that adequacy.
parameter (Table 9). Based on this analysis, conditions

Table 8. Comparison of design ESALs (UCB fatigue vs. Caltrans)

Traffic Base type Subgrade Caltrans ESALs ESAL ratio (UCB fatigue system to Caltrans
index R-value design system)
Santa Daggett Blue Canyon
Barbara
7 ctb 5 131,171 0.76 0.98 1.03
ctb 20 131,171 0.85 1.10 1.16
ctb 40 131,171 1.12 1.45 1.52
ab 5 131,171 0.43 0.55 0.58

ab 20 131,171 0.49 0.63 0.67

ab 40 131,171 1.05 1.22 1.39

11 ctb 5 5,565,306 0.38 0.37 0.48

ctb 20 5,565,306 0.64 0.62 0.81

ctb 40 5,565,306 I .83 1.75 2.30

ab 5 5,565,306 1.21 1.15 1.52

ab 20 5,565,306 1.13 1.09 1.43

ab 40 5,565,306 1.55 1.48 1.95

15 ctb 5 72,843,811 0.82 0.76 1.02

ctb 20 72,843,811 0.27 0.26 0.34

ctb 40 72,843,811 2.07 1.93 2.59

ab 5 72,843,811
8 2.11 1.95 2.64

ab 20 72,843,811
8 0.55 0.52 0.69
ab 40 72,843,811 8 0.95 0.89 1.19
Table 9. Effect of study parameters on median ESAL ratio
Parameter Value Median ESAL ratio Parameter Value Median ESAL
ratio
f
7 0.95 Santa Barbara 0.87
Traffic index 11 1.25 Location Daggett 0.99
15 0.88 Blue Canyon 1.13
5 0.82 Aggregate 1.17
Subgrade R- 20 0.69 Base type Cement treated 0.91
value
40 1.56

MIX DESIGN, CONSTRUCTION (Figure 5). Two points are significant. First, at the design
SPECIFICATIONS, AND QUALITY ASSURANCE asphalt content of 4.9 percent (corresponding to a
stabilometer value of 35), the air-void content is
California mix-design and pavement-construction approximately 5.4 percent. Field compaction meeting the
practices have served extremely well for a very long 95-percent relative density requirement would produce
period of time. However, they apparently produce acceptable mixes with air voids up to about 10 percent,
pavement surfaces with relatively small asphalt contents certainly a level sufficiently large for concern about
and relatively large air-void contents. Unfortunately, as fatigue cracking. Second, these test results demonstrate
shown by the test results reported herein, such mixes may how, for a given level of compactive effort, the air-void
have marginal fatigue resistance. Because California content increases with a decrease in asphalt content. The
design practice doesn’t explicitly treat fatigue distress, use effect of compaction in the field is similar to that in the
of relatively dry, harsh mixes may provide opportunity laboratory. Thus any reduction in asphalt content,
for the development of premature fatigue cracking. This whether by intent or by construction variability, will
section briefly focuses on some of the implications of increase air-void content unless compensated by
current California design and construction practice. increasing the compactive effort. Degradation of fatigue
Field compaction of asphalt concrete is currently performance is the inevitable result.
specified in terms of relative density, that is, the ratio of To illustrate effects of asphalt and air-void contents
in-situ density to the density of laboratory specimens on fatigue performance, laboratory-fatigue-life estimates
compacted at the design asphalt content. The Triaxial were made based on strains in the 18 hypothetical
Institute kneading compactor is used for laboratory pavement structures for 15 different mixes (five asphalt
compaction, and typical specifications require a contents and three air-void contents) using stiffness and
minimum relative density of 95 percent. The relative fatigue measurements taken earlier (Harvey et al., 1996).
density specification allows quite large, in-situ air-void Fatigue life was found to be particularly sensitive to
contents. For example, even with maximum laboratory air-void content (Figure 6). For example, a mix targeted
compaction (corresponding to the 4.0-percent minimum at 5-percent asphalt and 5-percent air voids will suffer a
air-void content allowed by the mix design procedure), 30-percent reduction in fatigue life if the air-void content
the minimum relative density requirement permits an in exceeds its target by 1 percent but only a 12-percent
situ air-void content of 8.8 percent. Recent testing at reduction if the asphalt content is shy of its target by
UCB of a number of field cores from accepted Caltrans 1 percent. If quality control and assurance testing during
projects reveals air-void contents typically within the construction permitted a l-percent deficiency in asphalt
range of 6 to 10 percent (Benson 1995). Larger values and a 3-percent excess of air voids, the combined effect
are sometimes obtained when the asphalt content is would be a quite significant 70-percent reduction in
reduced below that suggested by stabilometer testing, or fatigue life.
when a method specification is used to control In addition to mix-design and construction-
construction. specification practices, the effect of construction
Stabilometer test results for the Valley asphalt- variability is also of interest. To briefly illustrate this
Watsonville granite mix provide a useful illustration effect, attention first focused on one of the
e StabiIometer
Value
l Air-Void
Content
z 25
>
20 - Poly.
5
z 15 (Stabilometer
f IO
Value
0 - - - Poly. (Air-
c
a Void
.I

Content)
4 4.5 5 5.5 6 6.5

Asphalt Content (%)

Figure 5. Stabilometer test results for the Valley asphalt-Watsonville granite mix
L

I I I 1
I

4 4.5 5 5.5 6
Asphalt Content (%)
l 2% Air Voids l 5% Air Voids
A 8% Air Voids - I - I I - Linear (2% Air Voids)
I I - - Linear (5% Air Voids) - Linear (8% Air Voids)

Figure 6. Effect of asphalt and air-void contents on simulated fatigue-life ratio

18 structures previously identified, that one utilizing attributable to construction variability. Monte Carlo
aggregate base, resisting a traffic index of 11, and simulation was used to determine the l0th-percentile
supported on a subgrade having an R-value of fatigue lives resulting from these assumptions.
20 (11 ab20). Mix-proportion targets were assumed to be Without variability, the fatigue life was estimated to
5 percent for both asphalt and air-void contents. In-situ be approximately 15,300,000 repetitions (unadjusted by
asphalt and air-void contents were assumed to be temperature conversion factor, shift factor, etc.). As
normally distributed about these targets, and standard expected, construction variability was found to reduce
deviations were selected to represent ranges reasonably the 1 Otli-percentile life, quite significantly for large
standard deviations in air-void content (Figure 7). RICH-BOTTOM PAVEMENTS
Variability in asphalt content was a much less significant
factor than variability in air-void content both because From a prior mix performance evaluation (Harvey et
asphalt content can be more accurately controlled during al., 1996, and Harvey and Tsai, 1996), it is apparent that
construction and because fatigue life is less sensitive to decreased air-void content and increased asphalt content
asphalt content than to air-void content. result in increased fatigue life. Pavement structural
One additional simulation was performed to design should take advantage of these properties
demonstrate the effects of construction variability on wherever possible to improve the performance of the
fatigue performance. Pavement section 1 lab20 was again pavement. For most pavement structures and typical
used but with asphalt and air-void contents targeted at 5 traffic loads, fatigue cracking is assumed to begin at the
and 8 percent, respectively, and with construction bottom of the asphalt concrete layer, where tensile strains
variability as identified in Table 6. The fatigue life at the are usually largest. Here large asphalt contents and low
targeted asphalt and air-void contents is approximately air-void contents would be most beneficial. The potential
4,400,000. Simulation produced a median fatigue life for rutting, on the other hand., is usually greatest at the
approximately equal to this target: the mean fatigue life top of the pavement, within 100 mm (4 in.) of its surface.
and its standard deviation were approximately 5,400,000 Therefore, mixes used near the pavement surface must be
and 3,900,000, respectively. The simulated distribution more rut resistant, which usually means lower asphalt
of fatigue life, showing a distinct skew toward larger contents than might be permissible farther from the
lives, is shown in Figure 8. This example demonstrates pavement surface.
the use of Monte Carlo simulation in quantifying the As an example of potential improvements in
fatigue life distribution resulting fi-om construction pavement performance that might be obtained by
variability. Knowledge of the fatigue life distribution is including mix design information in the pavement design
useful to the highway agency in developing rational process, the fatigue life of pavements with larger asphalt
construction performance requirements and in contents in the bottom lift was evaluated using the mix
determining appropriate performance-related pay fatigue data and pavement structural designs developed
schedules. It is useful to the highway contractor in as part of this study. The resulting pavement designs
evaluating alternate ways to minimize contractor risk were termed rich-bottom pavements.
either by improving construction quality or by enhancing
either the mix or the structure to compensate for Designs for Rich-Bottom Pavements. The 1 8
construction variability.
hypothetical pavement structures described earlier served
Conclusions from the above analysis include the
as the basis for evaluating the rich-bottom-pavement
following:
concept. However, because they incorporated total
1) Specifying in-situ compaction by a relative
asphalt concrete thicknesses less than 150 mm (6 in.), the
density requirement is an ineffective technique for
six structures designed to resist traffic characterized by a
controlling fatigue distress because it permits relatively
traffic index of 7 were not used in the evaluation. Rich-
large air-void contents that may be detrimental to fatigue
bottom designs were created for the remaining structures
performance;
by replacing the bottom 50 mm (2 in.) of the asphalt
2) In-situ fatigue performance can be quite
concrete layer with a larger asphalt-content, smaller air-
sensitive to construction variability and, by inference, to
void-content mix of the same [Link] resulting
the caliber of the quality assurance program;
pavement structural designs are shown in Table 10.
3) To restrict the risk of premature failure, mix
Referenced to the original mix, each replacement
design must recognize and, if possible, compensate for
mix had an asphalt content 0.5 percent larger and an air-
expected construction practice; and
void content 3.0 percent smaller. For example, if the
4) T h e ability to quantify fatigue-life original mix had an asphalt content of 5.0 percent and an
distributions has ‘significant potential for use in air-void content of 5.0 percent, the bottom 50 mm of the
construction in a) establishing rational performance
asphalt concrete layer was changed to have an asphalt
requirements, b) taking actions to reduce the risk of
content of 5.5 percent and an air-void content of
failure to meet performance requirements, and
2.0 percent, while the remainder of the asphalt concrete
c) establishing performance-based contractor pay
layer remained unchanged. For this reason, besides being
schedules.
16,000,000
14,000,000
12,000,000 -+- Standard Deviation of
e Asphalt Content, 0%
s 1 o,ooo,ooo
Q,
3 8,000,000
.a,
% 6,000,OOO
LL - a - Standard Deviation of
4,000,000
Asphalt Content, OS%
2,000,000
0
0 0.5 I 1.5
Standard Deviation of Air-Void Content (%)

Figure 7. Illustrative effect of construction variability on fatigue life

0.9
0.8
0.7
E
_ 0.6
.I
.I

Q
2 0.5
0
6 0 .4

0 2,000,OOO 4,000,OOO 6,000,OOO 8,000,OOO 10,000,000 12,000,000 14,000,000

Simulated Fatigue Life (Without Shift Factor)

Figure 8. Dispersion of fatigue life resulting from construction variability (I lab20)

restricted to pavement designs for traffic index 11 and 15, designs was also used for the rich-bottom pavements,
upper layer asphalt contents of 6.0 percent and air-void except for the following:
contents of 2 percent were not included in the analysis. 1) The fatigue life of both asphalt concrete layers
was calculated using the maximum tensile strain at the
Predicted Performance of Rich-Bottom Versus bottom of each layer, and
Conventional Structures. The same procedure used to
calculate fatigue performance for conventional pavement
Table 10. Characteristics of rich-bottom pavement structures
Traffic Subgrade Layer Class B cement-treated base Class 2 aggregate base
index R-value
Thickness Stiffness Poisson’s Thickness Stiffness Poisson’s
(cm) (MPa) ratio (cm) (MPa) ratio
11 5 Surface 1 14.7 Varies 0.40 20.8 Varies 0.40
Surface 2 5.1 Varies 0.40 5.1 Varies 0.40
Base 19.8 221 0.30 15.2 172 0.45
Subbase 42.7 138 0.45 33.5 138 0.45
Subgrade 27 0.50 27 0.50
20 Surface 1 16.2 Varies 0.40 19.3 Varies 0.40
Surface 2 5.1 Varies 0.40 5.1 Varies 0.40
Base 16.8 221 0.30 15.2 172 0.45
Subbase 25.9 138 0.45 21.4 138 0.45
Subgrade 84 0.50 84 0.50
40 Surface 1 19.3 Varies 0.40 19.3 Varies 0.40
Surface 2 5.1 Varies 0.40 5.1 Varies 0.40
Base 15.2 221 0.30 15.2 172 0.45
Subbase
Subgrade 161 0.50 161 0.50
15 5 Surface 1 30.0 Varies 0.40 36.1 Varies 0.40
Surface 2 5.1 Varies 0.40 5.1 Varies 0.40
Base 15.2 172 0.30 15.2 138 0.45
Subbase 54.9 138 0.45 44.2 138 0.45
Subgrade 27 0.50 27 0.50
20 Surface 1 22.4 Varies 0.40 26.9 Varies 0.40
Surface 2 5.1 Varies 0.40 5.1 Varies 0.40
Base 25.9 221 0.30 18.3 138 0.45
Subbase 38.1 138 0.45 38.1 138 0.45
Subgrade 84 0.50 84 0.50
40 Surface 1 31.5 Varies 0.40 28.4 Varies 0.40
Surface 2 5.1 Varies 0.40 5.1 Varies 0.40
Base 15.2 172 0.30 15.2 138 0.45
Subbase 10.7 138 0.45
Subgrade 161 0.50 161 0.50

2) The smaller of the two calculated fatigue lives conversion factors, reliability multipliers, or shift factors.
was selected as the critical value for the pavement The increased fatigue life of the rich-bottom pavements
structure. is readily apparent from these figures. In m any case s the
The critical fatigue life in the rich-bottom pavements increase is quite large, approximating an order-of-
is plotted versus the fatigue life of the conventional magnitude improvement.
structures in Figures 9 and 10, for pavement structures These computations not only underscore the possible
designed for traffic indexes of 11 and 15, respectively. merit of rich-bottom designs, but they also illustrate the
The plotted values are based on laboratory measurements potential for including mix design information in
of fatigue life, without the application of temperature pavement structural design decisions.
1.0 E+8
Traffic Index =

U
1.0 E+5 1.0 E+6 1.0 E+7 1.0 E+8
Fatigue Life - Conventional Pavement

Figure 9. Fatigue-life improvement resultingfrom rich-bottom designs(TI of 11)

1.0 E+9

1.0 E+7

4.0 E+7 1.0 E+8 1.0 E+9

Fatigue Life - Conventional Pavement

Figure 10. Fatigue-life improvement resulting fromrich-bottom designs (TI of 15)

This potential can be realized by implementation of the necessary to move forward with decreasing construction
A-003A fatigue testing and analysis procedures used for air-void contents and increasing asphalt contents near the
this project. These procedures can provide information bottom of thick asphalt concrete layers.
l

MIX-DESIGN EXAMPLE Instead estimates were made based on regression

equations from the mix design experiment, described
Described in this section is an example illustration of previously (Harvey et al., 1996). Enriching and
the use of the fatigue mix design and analysis system. In densifying the mix was estimated to increase its stiffness
this example, a new’highway is being constructed, and its by a factor of approximately 1.15 and to increase its
pavement must accommodate 8,OOO,OOO ESALs in the fatigue life by a factor of approximately 2.18.
design lane during the design period. Climatic conditions Results of the extended analyses are summarized on
at the site are considered to be similar to those at Blue Figures 11 and 12. Figure 11 shows that the 24.1 cm
Canyon. Because of the importance of this facility and (9.5-inch) surface thickness remains inadequate for
the difficulty of maintaining traffic operations during 8,000,000 ESALs even at reliabilities below 80 percent.
resurfacing, a mix-design reliability of 95 percent has To preserve the desired 95-percent reliability, the surface
been targeted. A trial mix, consisting of Valley asphalt thickness must be increased fiom 24.1 to 29.2 cm (9.5 to
and Watsonville-granite aggregate, has been selected. 11.5 inches). Another alternative, the rich-bottom design,
The asphalt content has been set at 4.5 percent: to reduce nicely accommodates approximately 8,000,000 ESALs
the likelihood of rutting distress, it cannot be increased. with the original 24.1 cm (9.5-inch) surface thickness,
The in-situ air-void content achieved by normal Figure 12. Economics, feasibility, and engineering
construction techniques is expected to be, 8 percent. judgment would, of course, dictate the final design
To ascertain whether this mix will provide adequate choice.
resistance against fatigue cracking, flexural-beam fatigue
tests were performed. Six specimens were tested, three at SUMMARY AND CONCLUSIONS
each of two strain levels. The average initial stiffness was
determined to be 6.84 GPa (992,500 psi), and the fatigue Described herein is a reliability-based mix design
life-strain relationship was quantified by regression and analysis system for mitigating fatigue distress in
analysis. The pavement structural design thicknesses and asphalt pavements. Design decisions consider not only
elastic parameters are as follows: fundamental mix properties but also the level of design
Layer Thickness Modulus Poisson’s traffic, the temperature environment at the site, the
(cm) (MPa) ratio pavement structural section, laboratory testing and
construction variabilities, and the acceptable level of risk.
Surface 24.1 6,845 0.40 Of particular significance is the integration of mix and
structural components and the explicit treatment of both
Class 2 15.2 172 0.45 testing and construction variabilities. The mix design and
aggregate base analysis system has been calibrated to generally conform
Subbase 21.3 138 0.45 with California experience and pavement design practice.
This calibration illustrates techniques by which the
Subgrade 84 0.50 theory and testing of the analysis system can be easily
adapted to any specific region. An example illustrates
Computations, summarized in Table 11, revealed how the system can be applied to a hypothetical design
that, in this application, the trial mix could withstand task.
approximately 2,800,000 ESALs at 95-percent reliability. Specific investigations reported herein lead to the
Because design ESALs totaled 8,000,000, adjustment following conclusions:
was necessary. Although available options included the 1) Construction control is an important
use of asphalt modifiers and, indeed, the use of entirely consideration in mix design. With respect to fatigue
different asphalts and aggregates, interest focused on performance, accurate control of air-void content is much
determining the effect of increasing the asphalt-concrete more important than accurate control of asphalt content.
thickness or using a rich-bottom design instead. Complicating this matter is the likelihood that smaller-
Examining the effects of less reliable designs was also of than-specified asphalt contents will result in increased
interest. air-void contents unless compactive effect is increased to
In the rich-bottom design, the bottom 51 mm compensate.
(2 inches) of asphalt concrete were replaced with an 2) In-situ fatigue performance can be quite
enriched (5-percent asphalt), well compacted (5-percent sensitive to construction variability and, by inference, the
air voids) mix. In determining properties of this mix, caliber of the quality assurance program. To restrict the
laboratory test data were assumed to be unavailable. risk of premature failure to tolerable levels, mix design
Table 11. Summary of parameter calculations in mix-design example
Parameter Source Notes
ESALs N&SF
TCFdi
N .
1.364 .l0-8 E- 3’412 Calibration from laboratory testing

e 0.0000672 ELSYM5

SF

TCF 1.448 In($) -2.475 Blue Canyon calibration, d = thickness of asphalt concrete in mm

M Z = 0.84 1, 1.28, and 1.64 for reliabilities of 80, 90, and 95

percent, respectively
1.093 Monte Carlo simulation assuming that laboratory testing
included three replicates at each of two strain levels
(150 and 300 microstrain)

and/or structural design must recognize and, if possible, specifically a R-value of 20; c) designs for coastal
compensate for expected construction practice. regions, specifically Santa Barbara; and d) designs
3) Rich-bottom designs, in which the bottom incorporating class B cement-treated base.
portion of the asphalt surface course is enriched with
added asphalt and compacted with fewer air voids, have REFERENCES
the potential for enhancing fatigue performance,
extending fatigue life by up to an order of magnitude Bell, C., A. Wieder, and M. Fellin (1994), “Laboratory
beyond that of conventional structures. The investigation Aging of Asphalt-Aggregate Mixtures: Field
of rich-bottom pavements is an example of the power of Validation,” Strategic Highway Research Program
the mix design and analysis system to quantitatively Report No. SHRP-A-390, National Research
explore materials and structural alternatives that take Council, Washington, D.C.
advantage of improved fatigue properties at critical Benson, P. (1995), “Comparison of End-Result and
locations in the pavement structure. Method Specifications for Managing Quality,” A
4) Monte Carlo simulation is an effective paper presented at the 74th annual meeting of the
technique for simulating fatigue-life distributions Transportation Research Board, Washington, D.C.
resulting from testing, extrapolation, and/or construction Deacon, J., J. Coplantz, A. Tayebali, and C. Monismith
variabilities. The ability to quantify fatigue-life (1994a), “Temperature Considerations in Asphalt-
distributions has significant potential for a) establishing Aggregate Mixture Analysis and Design,”
rational performance requirements, b) taking actions to Transportation Research Record 1454,
reduce the risk of failure to meet performance Transportation Research Board, pp. 97-112.
requirements, and c) establishing performance-based Deacon, J., A. Tayebali, J. Coplantz, F. Finn, and
contractor pay schedules. C. Monismith (1994b), “Fatigue Response of
5) The mix design and analysis system can be Asphalt-Aggregate Mixes, Part III-Mix Design and
effectively used to determine the consistency of structural Analysis,” Strategic Highway Research Program
design procedures with respect to the control of fatigue Report No. SHRP-A-404, National Research
distress. For example, conditions where California Council, Washington, D.C.
practice is most likely to yield designs that are vulnerable Dempsey, B., W. Herlache, and A. Pate1 (1985),
to fatigue cracking appear to include a) designs for large “Volume 3. Environmental Effects on Pavements-
levels of traffic loading, specifically a TI of 15; Theory Manual,” FHWA/RD-84/115 University of
b) designs for subgrades of intermediate strength, Illinois at Urbana-Champaign.
40,000,000

35,000,000

30,000,000

25,000,000

y 20,000,000
<
$ 15,000,000

10,000,000

5,000,000

0
9 9.5 10 10.5 11 31.5 12
Thickness of Asphalt Concrete (in)
+ - - 80-Percent Reliability - -w - 90-Percent Reliability
I I

- 95-Percent Reliability
Figure 11. Effect of surface thickness and design reliability
on in-situ traffic resistance (conventional pavement)

40,000,000

35,000,000

30,000,000

25,000,000
cn
< 20,000,000
XI 15,000,000

1 o,ooo,ooo

5,000,000

0
9 9.5 I0 10.5 11 11.5 12
Thickness of Asphalt Concrete (in)
- _ + - - 80-Percent Reliability - -m - 90-Percent Reliability
_jt_l_ 95-Percent Reliability

Figure 12. Effect of surface thickness and design reliability

on in-situ traffic resistance (rich-bottom pavement)
Epps, J. (1996), “Accelerated Field Test of Performance- Program,” Strategic Highway Research Program
Related Specifications for Hot-Mix Asphalt Report No. SHRP-A-404, National Research
Construction-Task G, Interim Report,” submitted Council, Washington, D.C.
by the Nevada Automotive Test Center et al. to the Tayebali, A., J. Deacon, J. Coplantz J. Harvey, and
Federal Highway Administration, March, Chap. 12. C, Monismith (1994b), “Fatigue Response of
Granley, E. (1969), “Part 4-Variations of Bituminous Asphalt-Aggregate Mixes, Part I-Test Method
Construction,” Public Roads, Vol. 35, No. 9. Selection,“’ Strategic Highway Research Program
Harvey, J., J. Deacon, B. Tsai, and C, Monismith (1996), Report No. SHRP-A-404, National Research
“Fatigue Performance of Asphalt Concrete Mixes Council, Washington, D.C.
and Its Relationship to Asphalt Concrete Pavement
Performance in California,” RTA-65 W485-2,
DISCLAIMER
Asphalt Research Program, CAL/APT Program,
Institute of Transportation Studies, University of
California, Berkeley. The contents of this paper reflect the views of the
Harvey, J., and B. Tsai (1996), “Effects of Asphalt authors who are responsible for the information and the
Content and Air-Void Content on Mix Fatigue and accuracy of the data presented herein. The contents do
Stiffness,” to be published by the Transportation not necessarily reflect the official views or policies of the
Research Board. California Department of Transportation or the Federal
Herlache, W., A. Patel, and B. Dempsey (1985), “The Highway Administration. The State of California does
Climatic-Materials-Structural Pavement Analysis not endorse products or manufacturers discussed in this
Program User’s -Manual,” FHWA/RD-86/085, report. Trade or manufacturer’s names appear herein only
University of Illinois at Urbana-Champaign. because those identifications are essential to the report.
Tayebali, A., J. Deacon, J. Coplantz, F. Finn, and This report does not constitute a standard, specification,
C. Monismith (1994a), “Fatigue Response of or regulation.
Asphalt-Aggregate Mixes, Part II-Extended Test