TRANSPORTATION RESEARCH RECORD 1207 121

Rigorous Application of Linear Damage

Concepts in Development of Improved
Flexible Pavement Performance Models
STEPHEN B. SEEDS AND LUIS M. MEDUS

This paper describes the development of improved flexible most obvious application of a damage model is in the pave-
pavement performance prediction models in which Miner's ment structural design process where it provides a means for
linear damage hypothesis was rigorously applied in evaluating the determination of pavement layer thicknesses. Depending
original data from the AASHO Road Test. Effects of seasonal on the nature of the model, it also provides a basis for deter-
variation of soil and pavement properties were considered along mining the relative effects of different wheel loads, tire pres-
with the actual steering and trailing axle loads within the linear sures, and load configurations on a pavement's load-carrying
damage framework of Miner's hypothesis. Separate models to capacity. The latter provision translates further into a means
predict the number of single- and tandem-axle loads sustained for converting mixed-axle-load traffic into an equivalent design
were developed using five mechanistic response criteria: asphalt number of axle load repetitions of a uniform magnitude.
concrete (AC) tensile strain, AC tensile stress, AC shear strain,
Existing damage models vary from empirical (relying on
AC shear stress, and roadbed soil vertical strain. The single-
and tandem-axle models based on AC tensile strain had the experience or observation alone) to mechanistic (relying on
highest overall precision, i.e., coefficients of determination (r2) engineering mechanics). Historically, pavement performance
of 0.83 and 0.68, respectively. The models correlate highly models have been empirically derived; however, there is now
with Road Test data, but they do not compare well with other a trend toward developing mechanistic-empirical models. These
performance models or even the basic AASHO Road Test per- models are based on mechanistic response factors (i.e., stress,
formance equation. The implication is that the improved models strain, and deformation) but are statistically calibrated to
require their own set of standard 18-kip equivalency factors observed performance.
for use in projecting the number ofload applications that would Existing pavement damage models have one of two general
be used in designing a flexible pavement structure. criteria for failure: one is pavement condition (i.e., extent
and severity of distress); the other is pavement roughness
A study for the Arizona Department of Transportation (DOT)
(i.e., ride quality or serviceability). The AASHTO flexible
was recently completed to evaluate increased pavement load-
pavement performance algorithm (1) is an example of an
ing. In it, new procedures were developed for accurately con-
empirical damage model having terminal serviceability as its
sidering the effects of load magnitude, load configuration,
failure criterion. Fatigue damage equations developed under
and tire pressure on pavement design and performance. One
NCHRP Project 1-lOB (2) are examples of mechanistic-
major task in that study was the development of improved
empirical models having an allowable level of cracking as their
mechanistic-empirical models to simulate the performance of
failure criterion. In general, empirical models are adequate
flexible pavement sections at the AASHO Road Test. The
for predicting future performance under conditions similar to
models, which are based on a rigorous application of elastic-
those under which observations are made; however, they are
layer theory and Miner's linear damage hypothesis, were used
not necessarily reliable for predicting performance under con-
to develop an improved set of load equivalence factors and
ditions outside those inherent in their development. Mechan-
a new mechanistic pavement design (McPAD) procedure.
istic (or mechanistic-empirical) models are better suited for
The focus of this paper, however, is on model development.
prediction outside the range of the data from which they were
developed, since they rely on pavement responses generated
by proven theoretical models for their extrapolation. Because
BACKGROUND
of the need to consider loads and tire pressures significantly
higher than those considered in the past, a mechanistic approach
A damage-based pavement performance prediction model (or
was selected for developing the damage-based prediction models
damage model as it will sometimes be referred to) is an equa-
in this study.
tion that can be used to predict the number of load applica-
tions that can be sustained by a given pavement structure in
a given environment before it reaches a certain failure cri- CRITERIA FOR DAMAGE MODEL DEVELOPMENT
terion. (In this context, a damage model does not have to be
one that is based on fatigue cracking; it only has to be one In addition to the use of a mechanistic-empirical approach,
that considers cumulative load applications.) The primary and the following criteria were selected for the development of
[Link]-based pavement performance prediction models for
ARE Inc., 2600 Dellana Lane, Austin, Tex. 78746. Arizona DOT:
122 TRANSPORTATION RESEARCH RECORD 1207

1. AASHO Road Test data. Although 25 years old, the SEASON

data base from the original AASHO Road Test experiment 1958
OCT Fall
is still the best organized, most extensive and accurately col-
lected set of roadway performance data. NOV 19
2. Seasonal variation of roadbed soil support. To develop DEC
Winter - Frozen
a mechanistic damage model with a potentially higher degree 1959
JAN
of accuracy than that of previous research efforts, it was con-
sidered essential that the seasonal variation of roadbed soil FEB 25 + Winter - Frozen or
support at the Road Test be evaluated. To accomplish this, 6 Spring - Thaw
MAR Spring - Thaw
it was necessary to translate seasonal deflections and labo-
ratory test results into pavement material properties so that APR

the resulting variation of critical pavement stresses and strains HAY

could be considered. Miner's linear damage hypothesis (3)
was assumed to be valid, thus allowing the individual seasonal JUN
Summer
damages for each AASHO Road Test section to be accu- JUL
mulated and used in the analysis process.
AUG
3. Consideration of the effects of steering axles independ-
ently from load axles. Since steering axle loads ranged as high SEP 23
as 12,000 lb at the Road Test, it was decided that their effects
OCT
should be considered separately from the trailing load axles. i;aJ_1

This was accomplished within the same linear damage frame- NOV
2
work used for considering seasonal effects. Winlar - Uorrozen
DEC
4. Serviceability as performance criterion. Traffic repeti- 30
1960 Winlar - Frozen
tions corresponding to a serviceability index of 2.5 were used JAN 23
Sprl!!!j - Thew
in developing the damage models. Traditionally, pavement FEB
10
damage has been associated with the development of crack- Winter - Frozen

ing; however, there was no reason not to associate it with loss MAR 23
of serviceability. Spring - Thaw
APR
5. Separate damage models for single- and tandem-axle 4
MAY
loads. This was included in the criteria for model development
in order to maximize precision and to provide a better basis JUN
for evaluating the relative difference between single- and tan- Summer
JULY
dem-axle loads.
AUG

ANALYTICAL PROCEDURE SEP 21

OCT
Fall
Several steps were accomplished in the process of developing
the damage models. These steps are discussed in a logical NOV
30
sequence below.
FIGURE 1 Seasonal divisions established for
AASHO Road Test experiment.
Step 1: Section Selection

All of the primary AASHO Road Test flexible pavement

sections consisted of cross sections having three pavement ings of NCHRP Project 1-lOB (2). These are depicted in
layers: asphalt concrete surface, granular base, and granular Figure 1. Note that because of the different rates of thawing
subbase. In choosing sections for detailed analysis, only those associated with section thickness, there is a variable division
meeting the following layer thickness constraints were included: between the first hard-freeze winter and the first spring thaw.
asphalt concrete surface thickness (D 1 :::: 2 in.), base thickness This variation was handled on a section-by-section basis. The
(D 2 :::: 6 in.), and subbase thickness (D 3 :::: 8 in.). These layer asphalt concrete elastic modulus values for each of the seasons
thickness constraints were selected in order to confine the were based on laboratory test results and recommendations
analysis to sections having significant load-carrying capacity. from NCHRP 1-lOB (2): summer (230 ksi), fall (450 ksi),
Since several of these sections did not reach a terminal ser- wintPr lcoi'\ ~nri onrino (71 ()- lcoi'\
------ (1
'\-, 7()()
- - ----/ ----- -.-----o" -/
viceability of 2.5 during the two-year traffic loading per-
iod, only 33 single-axle and 27 tandem-axle sections were
considered. Step 3: Determine Cumulative Load Applications

Step 2: Season Delineation Appendix A of AASHO Road Test Special Report 61E (4)
was used to determine the cumulative number of wheel load
Primary seasonal divisions were established on the basis of a applications sustained by each section (single and tandem)
detailed examination of seasonal deflections and on the find- until it reached a serviceability of 2.5.
Seeds and Medus 123

Step 4: Determine Seasonal Deflections equivalent to the actual wheel loads (compare Table 2 with
Table 1), it was necessary to include this additional task as
The graph in Figure 2 provides an example of deflection vs. part of the materials characterization process. To predict sea-
time for Section 253. That plot represents the pavement sur- sonal material properties under the actual applied wheel loads,
face deflection under a 30-kip single-axle load measured using two additional computer programs were developed: STAX-1
a Benkelman beam. The plot indicates the critical deflection and TANDAX-1.
values that were selected for each season. Note that in one Since the loads used to measure deflection matched the
case (summer 1959), deflections were divided into two sub- actual single-axle wheel loads in Lane 1, STAX-1 was designed
seasons because of a significant difference in deflection at the only to estimate the material properties under the Lane 1
beginning and end of the season. This subdivision was con- steering-axle loads. The diagram in Figure 4 shows roadbed
sidered necessary because of the potential impact on materials soil resilient modulus vs. deviator stress, which illustrates this
characterization and was applied on several other sections. process for a given section during a given season. The solid
Seasonal deflection estimates were made for all sections line is established by plotting the modulus-deviator stress
under the different deflection loads. Table 1 identifies the values generated in Step 5 for the two deflection loads. (The
single-axle loads that were used to measure deflection on each slope of the line that connects these two points is indicative
of the sections. Recall that Lane 1 was loaded solely with of the roadbed soil's sensitivity to load.) The theoretical steer-
single-axle load groups whereas Lane 2 was loaded [Link] ing-axle relationship was generated by solving for the deviator
with tandem-axle load groups. (The tandem-axle trucks did stress values corresponding to the two previous roadbed soil
have single-axle steering axles.) modulus values. In solving for these deviator stress values, it
was still necessary to satisfy the bulk stress criteria for the
base and subbase materials. The intersection of the two lines
Step 5: Characterize Seasonal Material Properties Under defines the point at which roadbed soil stress conditions under
Deflection Loads the steering axle are consistent with the in-situ behavior of
the soil. Thus , the roadbed soil resilient modulus and cor-
To characterize the seasonal material properties of each sec- responding base and subbase modulus values at this point
tion, a computer program, MODEST-1, was developed which represent the material properties required for steering-axle
basically uses an elastic-layer-theory model, ELSYM5 (5), to load conditions.
identify a unique set of pavement layer moduli that will match The actual stresses and strains for each season of each seg-
the specified critical seasonal deflections and satisfy the bulk ment are a by-product of the MODEST-1 and STAX-1 solu-
stress relationships established in NCHRP Project 1-108 (2) tions. The results for this and the previous five steps [as they
for base and subbase materials at the Road Test. Figure 3 pertain to all single-axle (Lane 1) sections] are included within
provides a flow diagram of the basic iterative deflection matching the draft final report to the Arizona DOT (6), but are too
process used by the MODEST-1 program. The tolerances lengthy to include in this paper.
selected for satisfying the bulk stress and deflection criteria
were 5 and 3 percent, respectively.
Step 7: Solve for Seasonal Material Properties for Each
Step 6: Solve for Seasonal Material Properties for Each Lane 2 Section
Lane 1 Section
Unlike the single-axle sections, neither the steering- nor the
Unfortunately, because of stress (or load) sensitivity of the tandem-axle loads in Lane 2 matched the loads used to mea-
materials and the fact that the deflection loads were not always sure deflection. Consequently, it was necessary to incorporate
a slightly different approach into the TANDAX-1 program
to solve for the material properties required for the Lane 2
TABLE 1 AXLE LOADS USED IN AASHO ROAD TEST sections. As can be seen in Table 1, most of the deflections
DEFLECTION STUDIES (4) in the Lane 2 sections were measured using only a 12-kip
single axle. Thus , to solve for the material properties under
Loop Lane Single Axle
Load (kips) the steering- and tandem-axle loads, the single resilient mod-
ulus vs. deviator stress point (derived from the MODEST-1
program for the 12-kip single-axle load) had to be combined
2 1 with an estimate of the roadbed soil's stress sensitivity.
2 6 The single resilient modulus vs. deviator stress point from
6, 12 the MODEST-1 deflection analysis is plotted, and a straight
3 1
2 6 , 12 line corresponding to the estimated stress sensitivity (slope)
is drawn through the point. This line represents the in-situ
4 1 12, 18 resilient modulus vs. deviator stress relationship for that sec-
2 12 tion during that season. Since the stress sensitivity for most
of the Lane 2 sections was unknown, individual estimates were
5 1 12, 22 . 4
2 12 based on the calculated stress sensitivity of the adjacent Lane
1 sections. For the cases where Lane 1 information was
6 1 12, 30 unavailable, stress sensitivity estimates were based on trends
2 12 observed in other Lane 1 sections.
Once the in-situ relationship for the roadbed soil was estab-
 [ -· - - - - Inner WNel Petll
Fl11ible p, ......
ent Performance Record Thie•- 4 -
. 6 - 16 - - [Link]- No. HI - - ,
Looo No • - I

. ....
,,.._..., .,
-
ouw ....., ..., .. Let11tll_QQ__ f t . - - -
!!'I~- S•P'. I Oct.
.. ..... ( .-. I Fe•. I .... I•-. June J11I• • - 1-l"Oct ..... Dec. ..... ~· June
- ~
Jiii•
Aalt~ •A -·-
·id: - ~IQ£+- --
I

-- ·-· -- ··- - --.

[.I'
._. · ·:i-c ..
. I -- I ·- : --. I -1 T - --
I ,.,---.;.___: L_.----.:. - ..
• I
. ----- ...
lid ' --- -- ...
I - ... - - 1- - - ·~ -·
. -
' ---~ :
,,.--
I r I

j r l -.... : I
r 1 I

:-- I ~
.. - ·- -
__ I r

'J;E;.
,
""•1 r I

•o
!I 1,.
I
I
I
--'=
--. ....
I '
I
..!
-- ·- .. ·r · . - ·-
,.r
! CLASS Z aS
ii
Clt&QI- ~ [Link] • c•AOo- D I

I
c =1 S . .I ..ATCM
• _, 90
llD DUI' ..&TCM

...
.u
1; • OVl•LAT

! I 11·
•,ao
i
·n
••! I·.. -
I
oe,... OD - -
1Jf eo

.!

••
10;
f o:I .,

c
••
!
-: IO
ii
40

-.
0
-
-·-
.. . . _..

-·
I
-rM Ill .
-
.. ..

-
. . -

1 ~-
. ~-
c __
-- .- ..
.. - -
-
~ t fC. ... -~ --· -· -- - _. ... = c;;:;
~- -- . •r ~ ..
-
'-
. -· . ----

I ..
,,ji -'
-
-' .
,
0
--- - - -- - I
I
-
I i - -
I
Average C riticol I
.045 .000 .067 .065 I .049 .042 .044 .0 ,.. .05!5
Deflection Selected I
I
s EASON FALL WINT.- FROZEN SPRING . SUMMER FALL
I I III
~ SPRING

WINTER-UNFROZEN __./ \...._ WINTER-FROZEN

FIGURE 2 Example of average critical deflection values by season.

Seeds and Medus 125

HOOEST- 1
Elastic Modulus Estimates for
AASHO Road Test Hal.tri1ls

Enter structural layer thicknesses,

denection lo1d ind critical seasOl\lll
denection for section. Also enter initial
estimate or roadbed soil resilient modulus.

Assign AC elastic modulus according lo

season and NCHRP Project 1-108 results.

Estimate moduli for subbase and

base m1l.trl1ls using W1l.trw1ys
Experiment Stellon rel1tlonshlp.

Run ELSYHS. Solve ror m11e. deneclion

and bulk stresses In base and subbne
under deneclion lold.

Calculate new subbase and base

modulus estimates using rel1tionships
established in NCHRP Project 1-108.

Estimate new base

end subb1se moduli
by ever1ging
previous eslim1tes

Estimate new
modulus ror
roadbed soll .

Yes

END

FIGURE 3 Flow diagram of MODEST-1 program.

Jished, the theoretical steering- and tandem-axle relationships tions (6). Like the single-axle results, however, they are too
were generated and plotted in the same way that the steering- lengthy to include in this paper.
axle relationship was produced for the Lane 1 sections. Sim-
ilarly, the intersections of the two theoretical relationships
with that established for the in-situ soil represent the points Step 8: Develop Single-Axle Damage Models
at which the roadbed soil stress conditions under the given
steering and tandem-axle loads are consistent with the in-situ Separate damage models were developed for single- and tan-
behavior of the soil. Thus, the roadbed soil modulus values dem-axle loads. The reason for this was the desire to inde-
and their corresponding base and subbase moduli at these two pendently examine the effects of single and tandem axles. A
points are the material properties required for the two par- combined model would have required some assumption as to
ticular loading conditions. the relative impact on pavement performance of positioning
The actual stresses and strains for each season of each sec- two single-axle loads of a given magnitude in a tandem con-
tion are, in this step, a by-product of the TANDAX-1 solu- figuration. This assumption would have introduced an addi-
126 TRANSPORTATION RESEARCH RECORD 1207

TABLE2 TEST VEHICLE LOADINGS AT AASHO ROAD soil, [Link]· The first four of these mechanistic responses were
TEST (4) calculated at the bottom of the surface (asphalt concrete)
layer and were considered in order to determine if any one
WEIGHT IN KIPS
in particular is a better predictor of pavement performance

e···
LOOP LANE FRONT L~D GROSS
AXLE AXLE WEIGHT than the other. The last response, vertical strain at the top

{2)
Ci).,
[Link] ..!o 2 2 4 of the roadbed soil, was considered because of its appli-
cability in predicting the performance of thin-surfaced

F~ONT •
[Link]
2 8 8
pavements .
As discussed in Step 6, seasonal values for all the mechan-
istic responses were generated using the ELSYM5 program

~ {©fl.'!.
(5), based on elastic-layer theory. Actual values for each load
Lt:
®""
4 12 28 and season of each section are contained within the single-

FRONT LOAD •••

[Link]
8 24 54
axie data base presented in Appendix B of the report to
ADOT (6).
The machinery for producing the a0 , a,, and a2 coefficients

,, 8 18 42
for the damage models was incorporated into a program called
DAMOD-4. Figure 5 is a flow diagram of the major opera-
tions of this interactive program.
First, the desired mechanical response is identified and, for
••• II 32 73 a specified trial a2 value, initial values for both a0 and a, are
provided (operation 1). The program then goes through every

@ {®--= ,, 8 22.4 SI
season for a given section and calculates the allowable load
repetitions for both the steering- and the single-axle loads
(operations 2 and 3). The next two operations (4 and 5) require

•••
@
II 40 811
an explanation of a technique derived by Taute et al. (7) which
uses Miner's linear damage hypothesis (3) to consider multiple
seasons and nonuniform axle loads in developing a new dam-

{®--= •• II 30 811
age model.
The linear damage hypothesis basically implies that one
repetition of a given stress or strain produces the same amount
® @ •
....
••• 12 48 108
of damage to a pavement whether it is applied at the begin-
ning, middle, or end of the pavement's life. It can be expressed
mathematically as follows:

(2)
tional source of error into the analysis and also made it impos-
sible to use the model to examine the effects of axle
where, in this case,
configuration.
To apply a mechanistic analysis approach using elastic-layer D total damage to the i'h section,
theory and Miner's linear damage hypothesis, it was first nec- ni actual number of stress or strain repetitions of a
essary to assume a form for the damage model. Previous given load during a given season,
research efforts, including NCHRP Project 1-lOB (2), sug- (N1)i = allowable number of stress or strain repetitions of
gested a form which was adopted for this study: a given load during a given season, and
m = product of the number of different axle loads times
(1)
the number of different seasons (on the i'h section).
where, in this case,
The allowable number of repetitions, (N1)i, is determined by
N1 = estimated number of load repetitions to solving the damage model (Eq. 1) for the stress or strain level
serviceability of 2.5, corresponding to a given axle load and season. The key to
R = selected mechanistic response (i.e., stress estimating the a0 , a,, and a2 coefficients in the damage model,
or strain), then, is to find an effective stress or strain level that would
EAc = estimated elastic modulus of the asphalt produce the same amount of damage to the section as the
concrete, and combination of all the axle load repetitions during the· dif-
n-~ n. ~ :inn n_ = [Link]~ tn hP [Link] thrnneh ~t:i­ ferent seasons. This means that the total damage (to the i'h
tistical analysis of the data. section) can also be expressed as:
m
The mechanistic responses that were considered in devel-
oping damage models (for both the single- and tandem-axle 2:ni
D= ~ (3)
loads) include: (1) maximum asphalt concrete (AC) tensile NJ)orr
strain, EAc; (2) maximum AC tensile stress, aAc (psi); (3)
maximum AC shear strain, 'YAc; (4) maximum AC shear where (N1).n is the allowable number of load repetitions cor-
stress, '!'Ac (psi); and (5) maximum vertical strain on roadbed responding to the "effective" stress or strain level.
Seeds and Medus 127

,.-- theoretical steering axle relationship

1
I
6868

·;;;
0.
6716
VI
er:
ILi I / slope indica live of
.,;
:I
I f in-situ sensitivity to
I
3
'O
o~

~ .!
0
u I
I -..,.. stress (load)

.
-;:
="'
(/')

0
I
"'~
~~ 622 1
'ii
VI

..
-..
'O

~
0 'O
0
er:
.,>
0
.t::.

2..02 J 2.13 .51 4.33

2.10

Roadbed Soi l Oeviator Stress, <1d (psi)

(Log Scale)

FIGURE 4 Graph of roadbed soil resilient modulus vs. deviator stress,

illustrating technique used to solve for material properties under steering-
axle loads in Lane 1 sections.

Rearranging the 'terms to solve for (NJ)eff and recognizing Once effective stress or strain values are calculated for each
that the total damage is calculated using Equation 2 gives: section, a regression analysis (operation 8) is performed on
NJ (in this case, the actual number of load repetitions expe-
(4) rienced by the section before it "failed") vs. [Link] to generate
new a0 and a1 coefficients for the damage model. A measure
of the "fit" of the model to the data, known as the coefficient
of determination (or r2 ), is also generated as part of this
Substituting the form of the damage model for NJ in Equation regression analysis .
1 and solving for the effective stress or strain, Rem results in: Operation 9 provides a test of whether the new a0 and a 1
values are significantly different from the assumed initial val-
ues. If they are, then the process must be reexecuted using
the new a0 and a1 values as initial estimates (operation 10).
(5)
When the a0 and a1 values are essentially equivalent to the
assumed initial values (operation 11), they represent the "best"
solution for the trial a 2 value .
Note that because asphalt concrete elastic modulus, EAc, is Table 3 is an example of output from the DAMOD-4 pro-
in the equation, it is necessary to calculate the effective stress gram for one of the initial asphalt concrete tensile strain models.
or strain, Rem for a modulus value corresponding to a par- For the trial a2 value of - 3.97, eight iterations were required
ticular season. Since it occurs between the extreme seasons, before the final a0 and a 1 values matched the initial specified
fall (autumn) was selected as the season for Reff calculations. values. These values, then, represent the best combination of
Thus, the asphalt concrete elastic modulus value used was a0 and a1 for the selected trial value of a2 • To get the best
450,000 psi. It should be recognized that the selection of fall combination of a0 , a,, and a2 , it was necessary to try different
as the season for Rerr calculations theoretically has no effect a2 values with the objective of finding the combination that
on the ultimate predictive accuracy of the damage model. provides the highest coefficient of determination (r2 ). Table
Operations 2, 3, 4, and 5 of the flow diagram in Figure 5 4 illustrates how the a2 value of -3.97 and the corresponding
are performed for one Road Test section at a time . Conse- a0 and a, values of 6.89 and - 6.21 (respectively) provided
quently, operations 6 and 7 are included to provide a means the maximum r2 • Therefore , they represent the best set of
for incrementing through each section. single-axle coefficients.
128 TRANSPORTATION RESEARCH RECORD 1207

DAMOD-4
Single Axle D1m11CJ9 Models ror
Flexible P8119m8nls

Identify response desired and enter

trial •2 and inilial estimates or
a 0 and a 1.

2 Reid characteristics ror 1th section .

3
For each season or the 1lh section,
calculate the allowable load repelilions
for both the steering and single axle loads
using the current a 0 , a 1• and a 2 values.

4 Calculate the seasonal damage due to each

1xle load by dividing the actual number or
seasonel load applications by the aiiowllbie .
The total d1m1ge for the ith section is
the sum or these individual damages

5 Use the total cumulative damage for

the ;th section lo estimate an
effective value for lhe response,
F\tfr. durino the fall season.

7
Increment Yes
1th seclion

No
8 Conduct 1 liner regression
analysis or log Hf' versus
109 Reff • generate new ao and
a I v1lues Ind determine r 2.

10 . . - -- - - ' - - - - .
No Rewind data and
iterate using new
a 0 and a 1 values.

FIGURE 5 Flow diagram of DAMOD-4 program.

The DAMOD-4 analysis for single-axle loads was per- modd based on vertical strain at the top of the roadbed soil,
formed considering five different mechanistic responses the analysis indicated an impractical and undue correlation
(including asphalt concrete tensile strain). The results are with vertical strains sustained during the winter. This was
summarized in Table 5. Figure 6 illustrates how well the ten- probably due to the fact that the underlying materials were
sile strain model fits the Road Test data . However, this and assigned modulus values based on engineering judgment of
the other relationships were all considered initial or prelim- the properties during the winter rather than on the deflection-
inary single-axle damage models . Although they are certainly based materials characterization technique used for the other
valid and could be used for design or pavement performance seasons. Whatever the explanation, it was reasoned that if
prediction, additional equations (described next) were devel- pavement damage during the winter was indeed insignificant,
oped which may be more suitable. then a suitable damage model could be developed by not
In the process of trying to develop a single-axle damage considering the frozen-winter seasons in the DAMOD-4 anal-
Seeds and Medus 129

TABLE 3 SAMPLE DAMOD-4 OUTPUT FOR PRELIMINARY SINGLE-AXLE

LOAD MODEL

CRITICAL RE5KtlSE KR llAMl\GE MJDEL: AC, TENSil.E S'mAIN

'!RIALro. 8
Af) 6.890
Al -6.210
A2 -3.970

SEASCNAL EFl'ECl'S :
SHUN; -23. 2295000
SlttlER - 21. 2860600
FAIL -22.4432500
WINl'ER -24. 7348800

NSF.C D1 D2 03 'IRSlM DAMSlM 1RPRlM SIREF x y YRUM

111 2. 6. 8. .9315E-t-06 . 3528E-+Dl 264037. .000419 -3 .37760 28.41244 5.96918
155 3. 6. 8. . 9555E-+D6 .1587E-+Dl 602083. .000367 -3.43525 28.42348 5.98023
623 3. 6. 8. .1245E-t-06 .4273E-+DO 291340. .000413 -3.38449 27.53842 5.09517
601 3. 6. 12 . .9315E-t-06 .1739E-+Ol 535789. .000374 -3.42709 28 .41244 5.96918
577 4. 6. 8. .1140E-+D7 .1891E-+{)l 602813. .000367 -3.43534 28 .50016 6.05690
625 4. 6. 12 . . 1019E-+D7 .9352E-+DO 1089033 . .000334 -3.47670 28 .45122 6.00796
419 3. 6. 8. . 1095E-t-06 .3208E-+DO 341288 . .000402 -3.39555 27 .48267 5.03941
487 3. 6. 12 . . 1155E-t-06 .2264E-+DO 510210. .000377 -3.42367 27.50584 5.06258
471 3. 9. 8. . ll85E-t-06 .7254E-+DO 163350. .000453 -3.34402 27.51697 5.07372
455 4. 6. 8. . 1449E-t-06 .9820E-+DO 147563 . .000460 -3.33691 27.60432 5.16107
453 4. 6. 8. . l359E-+D6 .4430E-+DO 306764. .000409 -3.38809 27.57647 5.13322
425 4. 6. 12 . . 7575E-t-06 .1042E+-01 726904 . .000356 -3 .44843 28 .32264 5.87938
417 4. 9. 8. . 2520E-t-06 . 8012E-+DO 314520. .000408 -3 .38984 27.84465 5.40140
477 4. 9. 12 . . 1649E-+D7 .1429E-+<ll 1153994. .000331 -3 .48075 28 .66034 6.21709
469 5. 6. 8. . 8400E-t-06 . 7164E+-00 1172530. .000330 -3.48186 28.36753 5.92428
445 5. 6. 12 . . l052E-+D7 .6511E-+DO 1615064. .000313 -3.50426 28.46506 6.02181
303 4. 6. 8. . 1200E-t-06 .7707E-+DO 155697. .000456 -3.34067 27.52244 5.07918
323 4. 6. 12 . .1200E-t-06 .1071E+-Ol 112036 . .000481 -3.31765 27.52244 5.07918
253 4. 6. 16 . . 6765E-t-06 .1950E+-Ol 346877 . .000401 -3 .39669 28 .27352 5.83027
321 4. 9. 8. .1200E-t-06 . 1248E-+Dl 96156. .000493 -3 .30696 27 .52244 5.07918
267 4. 9. 12 . . 1620E-t-06 .1538E-+Dl 105324. .000486 -3 .31333 27.65277 5.20952
309 4. 9. 16 . . 1535E-+D7 .4167E+-Ol 368219. .000397 -3.40086 28.62922 6.18597
259 5. 6. 8. . 1365E-t-06 .6848E-+DO 199322. .000439 -3.35794 27.57839 5.13513
307 5. 6. 12 . .8775E-t-06 .1576E-+01 556653. .000372 -3.42976 28.38650 5.94325
305 5. 6. 12 . . 1890E-t-06 .6615E-+DO 285720 . .000414 -3.38312 27.71972 5.27646
327 5. 6. 16 . .1014E-+D7 .1700E-+<ll 596383 . .000368 -3.43459 28 .44929 6.00604
313 5. 9. 8. .6615E-t-06 .1551E-+Dl 426449 . .000388 -3.41113 28 .26378 5.82053
331 5. 9. 12 . . 8355E-t-06 .8997E-+DO 928599. .000342 -3.46555 28.36520 5.92195
325 6. 6. 8. .1530E-t-06 .4016E-+DO 380967. .000395 -3.40324 27.62795 5.18469
257 6. 6. 12 . . l070E-+D7 .1150E-+Dl 929806. .000342 -3.46564 28.47243 6.02918
263 6. 9. 8. . 7680E-t-06 . 8489E-+DO 904675. .000344 -3 .46373 28.32862 5.88536
271 6. 9. 8. .1058E-+D7 .1384E-+Dl 764358 . .000353 -3 .45194 28.46754 6;02428
311 6. 9. 12 . .1005E-+07 . 8740E-+DO 1149881. .000331 -3 .48050 28.44542 6.00217

REX:RESSICN LlNE IS :
Af) 6.883
Al -6.212
R-S11'\RE .599

Notes: NSEC = AASHTO section number; D 1 , D,, D 3 = layer thickness (in.) for surface , base,
and subbase; DAMSUM = total damage for section computed using a0 , a,, and a2 ; STREF =
effective fall stress or strain for section ; TRPRIM = allowable load applications corresponding to
effective fall stress or strain; X = log 10 of STREF (independent variable in the regression analysis);
YPRIM = log 10 of TRSUM (dependent variable in the regression analysis); Y = YPRIM minus the
fall seasonal effect.

ysis. When this analysis was performed, the results for the in damage that results in each section when the frozen winters
vertical strain model were so remarkable that similar analyses are included was indeed insignificant. This test basically con-
were carried out to develop models for the other four mechan- sisted of an examination of the differences between the dam-
istic response variables. The results are summarized in Table age calculated with the frozen-winter effects and those cal-
6 and a graph illustrating the relative precision for the asphalt culated without the frozen-winter effects. The results indicated
concrete tensile strain model is presented in Figure 7. that there was no significant difference for all 33 Road Test
A test of these models was made to determine if the increase section_s. Thus, it was concluded that the increase in damage
 TABLE 4 SAMPLE OF OPTIMUM COMBINATIONS OF a0 ,
a1 , AND a2 FOR SINGLE-AXLE DAMAGE MODEL

Coefficient of
Coefficients Determination

a2 al ao (r2)

-3.50 -6.46 3.33 0.588

-3.70 -6.35 4.85 0.597

-3.90 -6.23 6.40 0.597

-3.95 -6.21 6.78 0.597

-3.97 -6.21 6.89 0.599 (Optimum)

-4.00 -6.19 7.13 0.597

-4.10 -6.15 7.85 0.596

-4 , 30 -6 , 06 9_34 0.587

TABLE 5 INITIAL SINGLE-AXLE DAMAGE MODELS RESULTING FROM DAMOD-4

COMPUTER ANALYSIS

Mechanistic Optimum Coefficients Coefficient of

Response Symbol Determination
Considered (R) (r2)

Asphalt Concrete €AC 6.89 -6.21 -3.97 0.599

Tensile Strain

Asphalt Concrete (J'AC 4.68 -6.40 2.80 0.615

Tensile Stress

Asphalt Concrete '(AC 8.96 -6.43 -4.20 0.584

Shear Strain

Asphalt Concrete TAC 6.69 -6.28 2.10 0.562

Shear Stress

Vertical Strain fRS (Model not possible)

on Roadbed Soil

Form o f Da ma ge Model

log(Nf) - a 0 + a 1*log(R) + a 2*log(EAc)

 .0004 .0005
6.3

6.2 r--... r
....... l

6. 1

6 0 ""' " ~ D D
0
...,
0
~~ I .P 0
D
5.9 u

'-I'.
.........
Ul [ 0 Not< • ~ -. ,.. 1 ,t,nn ho n
c 5.8 '" m
0 F~ll eaac '1
:~ (j;;Al' 45( ,000 psi
+'
Cl)
a.
Cl)
5.7
' .......
'-...
a:: 5.6
01
0
_J
5.5
log Nt= 6.89 - 6 .21 *log t..Ac - 3.97 ll IOQ
"
EAC
r"...
~~ '"
5.4 ->-----
.........
I I
5.3 2
r =0.599
"'-.
0
'-...
5.2 u
0 D
' "'"'- n

5.1 LJ
oc D D
D
D
5 """' r--....
-3.51 -3.49 -3.47 -3.45 -3.43 -3.41 -3.39 -3.37 -3.35 -3.33 -3.31
Log Asphalt Concrete Tensile Strain

FIGURE 6 Single-axle damage model based on asphalt concrete tensile strain.

TABLE 6 SINGLE-AXLE DAMAGE MODELS RESULTING FROM DAMOD-4

COMPUTER ANALYSIS ON DATA WITHOUT FROZEN-WINTER EFFECTS

Mechanistic Optimwn Coefficients Coefficient of

Response Symbol Determination
Considered (R) (r2)

Asphalt Concrete £AC 3.25 -7.50 -4.10 0.834

Tensile Strain

Asphalt Concrete ()"'AC 2.69 -7.47 3.60 0.841

Tensile Stress

Asphalt Concrete YAC 6.61 -7. 72 -4.50 0.829

Shear Strain

Asphalt Concrete TAC 3.85 -7.62 3.10 0.819

Shear Stress

Vertical Strain £RS -7.75 -4.28 0. 723

on Roadbed Soil

Eo~m gt Damft ge Mogel

log(~f) - ao + a 1*log(R) + a 2*log(EAC)

132 TRANSPORTATION RESEARCH RECORD 1207

.0004 0005 .0006

6.2 ....
'-n
6.1 L..I
..........
6
5.9
5.8
5.7
0
1111
"" I][)"
c'
[](
lJ
n
0,...,
.... 0

..............
,...,
LI""
,_
"'"'
5.6

UJ
c:
0
5.5
5.4 '" ..............
Note• Reso ution based on
Fall Sea so
(EAr = 450 000 p ii).
:.:;
:.:;
5.3
r:::,.,
- ~
Cll ..J.,L
Q_ ~
Ql 5. 1
a:: ..............
5 ....
O'I ~
0 4.9 --
log Nf = 3. 25- 7.50 :flog tAC - 4.10 *log E AC 0

'-
....J I I I n
4.8 I I I 0 ..............
4.7 r2 = 0.834
4.6 .... -- ~
4.5
4.4
4.3 J
0
" .............
[~ 0 a
4.2
'
4.1
-3.5 -3.46 -3.42 -3.38
Log Asohalt Concrete Tensile Strain
-3.34 -3.3
r

-3.26 -3.22 ""

FIGURE 7 Single-axle damage model based on asphalt concrete tensile strain (frozen-winter effects not included).

due to load applications during the frozen-winter season is RECOMMENDED MODELS

indeed negligible.
Based on the results of the analyses, it was recommended
that the single- and tandem-axle damage models that are based
Step 9: Develop Tandem-Axle Damage Models on asphalt concrete tensile strain without frozen-winter effects
be used both for asphalt concrete pavement design and for
The approach to developing the tandem-axle damage models examining the relative effects of different loads, load config-
was almost identical to that for the single-axle models in Step urations, and tire pressures on pavements with asphalt con-
8. The form of the model was the same, the same five mechan- crete thicknesses greater than 2 in. The predictive accuracy
istic response variables were considered, and, for consistency, of all the models based on a mechanistic response in the
load applications during the frozen-winter season were not surface layer was very high; however, the single- and tandem-
considered. The principal difference in the analysis was in the axle tensile strain models had the highest combined precision.
recognition that damage due to the steering axles had to be The fact that most of the experience to date with asphalt
assessed using the appropriate single-axle damage model. The concrete damage models has been with tensile strain was another
necessary changes were incorporated into the DAMOD-4 reason for recommending these particular models.
program to produce DAMOD-5. The differences are in oper- The roadbed soil vertical strain models were only recom-
ations 1, 3, 4, and 5 (see Figure 5). mended for the case where surface treatments or thin asphalt-
In operation 1, fixed a0 , a1 , and a2 values from the single- concrete-surfaced pavements are being designed or evaluated.
axle model are entered along with the trial a2 value and initial Although these models have a somewhat lower level of pre-
estimates of a0 and a1 for the tandem-axle model. In operation cision, they still explain a high percentage of the variability
1. thP
-, ~llf'llur~hlP
---- ---- l"~rl --r---------
., ____ ------ --- thP
rPnPtltlnn1;;. fnr [Link] ~nil
---- --------o ------ t~nriPm-
---------- nh~PTVPrl in thP A_~SHO R0l!.d Ti:st dat:i..
axle loads are calculated using the appropriate a0 , a1 , and a2
values. In operation 4, total damage is calculated with par-
ticular attention to the load configuration (steering or tan- CONCLUSION
dem). In operation 5, the effective stress or strain is calculated
with a formula derived using the same basic approach as that In order to minimize the effects of extreme seasonal variations
used to derive Equation 5 for single-axle loads. observed at the AASHO Road Test and to provide a better
The final results of this step for all five mechanistic response basis for extrapolation to heavier loads, higher tire pressures,
variables are presented in Table 7. and more repetitions, a rigorous mechanistic approach was
Seeds and Medus 133

TABLE 7 TANDEM-AXLE DAMAGE MODELS RESULTING FROM DAMOD-5

COMPUTER ANALYSIS ON DATA WITHOUT FROZEN-WINTER EFFECTS

Mechanistic Optimum Coefficients Coefficient of

Response Symbol Determination
Considered (R) (r2)

Asphalt Concrete £AC 0.82 -6.18 -3.0 0.676

Tensile Strain

Asphalt Concrete 0-AC 0.91 -5.51 3.0 0.654

Tensile Stress

Asphalt Concrete YAC 5.19 -5.30 -3.0 0.580

Shear Strain

Asphalt Concrete TAC 4.75 -5.05 1. 9 0.578

Shear Stress

Vertical Strain [Link] -5.27 -3.42 0.649

on Roadbed Soil

Form of Damage Model

used to develop improved flexible pavement performance pre- REFERENCES

diction models. This has resulted in a methodology that should
be better suited for use over loads, tire pressures, and envi- 1. AASHTO Interim Guide for Design of Pavement Structures, 1972
ronments that are well outside those of the Road Test. (Chapter III Revised, 1981), American Association of State High-
way and Transportation Officials, Washington, D.C., 1981.
As part of the Arizona DOT study, revised load equivalency 2. F. Finn, C. L. Saraf, R. Kulkarni, K. Nair, W. Smith, and
factors and a computerized flexible pavement design proce- A. Abdullah. Development of Pavement Structural Subsystems.
dure (McPAD) were developed based on the new damage Vol. I. Final Report, NCHRP Project 1-lOB. TRB, National
models. Although the damage model coefficients (i.e., the a 1 Research Council, Washington, D.C., 1977.
values) are higher than those generated in past research stud- 3. M. A. Miner. Cumulative Damage in Fatigue. Journal of Applied
Mechanics, September 1945.
ies, comparisons with designs derived from the Road Test 4. Special Report 61E: The AASHO Road Test, Report 5, Pavement
performance equations indicate an excellent correspondence Research, HRB, National Research Council, Washington, D.C.,
for Road Test conditions. The differences occur when com- 1962.
parisons are made for loading and environmental conditions 5. G. Ahlborn. Elastic Layered System with Normal Loads. Institute
of Transportation Studies, University of California at Berkeley,
outside the Road Test, where heavier loads, higher tire pres- 1972.
sures, and stronger soils result in significantly different pave- 6. S. W. Hudson, S. B. Seeds, F. N. Finn, and R. F. Carmichael
ment structural requirements. Obviously, this means that the III. Evaluation of Increased Pavement Loading. Draft Final Report
final results of the study will need further investigation before by ARE Inc. for Arizona Department of Transportation, 1986.
the models and procedures can be implemented. 7. A. Taute, B. F. McCullough, and W.R. Hudson. Improvements
to Materials Characterization and Fatigue Life Prediction Methods
for the Texas Rigid Pavement Overlay Design Procedure. Research
Report 249-1. Center for Transportation Research, University of
ACKNOWLEDGMENTS Texas at Austin, 1981.

The authors thank the key Arizona DOT personnel, Larry

Schofield, Subodh Kumar, John Eisenberg, and Rich Powers,
for their support and direction on this project. They especially
thank the Principal Investigator, Fred Finn, for his technical Publication of this paper sponsored by Committee on Flexible Pave-
guidance and unwavering support during the rough times. ment Design.