Beruflich Dokumente
Kultur Dokumente
INTRODUCTION
Development of highway pavements that require minimum maintenance
during the in-service life is the primary goal of recent highway studies.
Estimation of pavement performance within acceptable levels is necessary
in selecting the best design for a given project.
Most design factors that affect pavement performance have some degree
of variation, which may come from dispersion of their values and/or errors
associated with estimating these factors. The uncertainty also arises from
the design models themselves. There are two sources of model error. First,
the number of basic physical variables has been limited to a finite number
in leaving out a possibly infinite set of parameters that in the model
idealization process have been judged to be of secondary or negligible
importance for the problem at hand. Second, model uncertainty is due to
the idealization of operational mathematical expressions. Besides this
cause of pragmatic simplification, uncertainty may be due to lack of
knowledge about the detailed interplay between the considered variables.
Historically, variation or uncertainty (in design) has been accounted for
by the use of safety factors or arbitrary decisions based on experience.
However, the use of safety factors, such as confidence levels in the design
of pavements with little consideration given to uncertainty of design
factors, has resulted in few failures (Hudson 1975).
In order to assess comprehensively the effects of uncertainty, probabi-
listic concepts need to be applied in an explicit, nonarbitrary way.
'Res. Asst., The Univ. of Mississippi, University, MS 38677.
2
Prof. of Civ. Engrg., The Univ. of Mississippi, University, MS 38677.
Note. Discussion open until October 1, 1988. To extend the closing date one
month, a written request must be filed with the ASCE Manager of Journals. The
manuscript for this paper was submitted for review and possible publication on July
17, 1986. This paper is part of the Journal of Transportation Engineering, Vol. 114,
No. 2, May, 1988. ©ASCE, ISSN 0733-947X/88/0002-0294/S1.00 + $.15 per page.
Paper No. 22448.
294
296
( START j
Hi f °EXP(-Z2/2)dz
( END J
297
Objectives
This study develops a simulation model for evaluating the reliability/
performance of pavements. The analysis relies on analytical formulation
relating performance and the basic design parameters. The model is
versatile in that it not only calculates the reliability of the pavement with
age, but also the expected or average life.
RELIABILITY/PERFORMANCE MODEL
The algorithm for calculating PSI values, pf, based on a set of input
design factors (independent variables) constitutes the first submodel.
When the dependent variable pf is a nonlinear function of independent
variables (concrete properties, soil strength, climatic factors, initial ser-
viceability index, as in JPC and CRC models) the mean and variance of pf
are more difficult to calculate, often done by numerical integration. The
point estimate method was employed first to perform the calculations. The
mathematical procedure, however, is very complicated because the design
equation is nonlinear and the application of two-point estimate may not be
accurate. The Monte Carlo simulation method, however, turned out to be
more practical and straightforward to use than the point estimate method.
The design equation model or the mathematical formulation embedded in
the design model is solved by the simulation technique in conjunction with
the assumed probabilistic formulations for all the input design factors.
298
In the model, all the design parameters including initial and terminal
serviceability indices are assumed probabilistic quantities (see Appendix
II). Relying on the results of previous studies (Hudson 1975), and for the
sake of simplicity, the writers assume that the input design factors are
normally distributed. The values of design factors are generated indepen-
dently by using a subroutine based on the central limit theorem. The
following algorithm has been developed for this purpose:
Making use of one set of selected input design factors (a single value
randomly selected for each factor) apy-value is computed using the design
model. The calculations are repeated 500 times, resulting in an equal
number of pj-values; each value results from a unique set of input design
factors. The mean and standard deviation of the 500 /^-values are now
calculated as a part of the submodel.
In order to verify the normal distribution assumption, the /^values,
computed by RAPP-I for a JPC pavement at an age of 10 years (design life
of 20 years) are plotted in a histogram in Fig. 2. Typical input data (the
mean value and coefficient of variation) for the RAPP-I model with
premium as well as AASHTO Guide equations are included in Appendix II,
Table 3. Note that the data fall under four general categories: traffic,
material properties, environment, and limiting criteria for serviceability.
The probability distributions of pyfor both premium and AASHTO Interim
Guide models are included in the figure. For all practical purposes, the two
histograms resemble normal probability functions—the histogram of the
premium model more so than that of the AASHTO model.
Model Uncertainty
The (uncertainty) analysis described in the preceding section ignores
model uncertainty or error due to imperfect formulation of the design
model. To estimate the model uncertainty one must in principle be able to
observe many differences between (or ratios of) observed and predicted
outputs. The observed values can be compared with model predicted
values to yield data from which estimates of standard deviation of model
prediction <sME, can be made. The aggregate standard deviation (error) of
the model, apf, must now be computed by combining, in accordance with
Eq. 6, <rME, and the error due to uncertainty in the design factors, vpd.
orp/=(a^ + a^)1/2 (6)
The model error of FLEX and JPC models (AASHTO Interim Guide
1972) are determined to be normally distributed with a standard deviation
of 0.24 and 0.46, respectively (Shah et al. 1985). The error associated with
the premium model for design of JPC pavements is also normally distrib-
uted with a standard deviation of approximately 0.15 (George et al. 1988).
The standard deviation, upf, with age as well as with serviceability level
299
40
(a)
20'
_, I iliMJ U""H •
0.0
—f—
2.0 3.0 4,0
—,
1.0
Serviceability Index 5.0
40.
(b)
20
0
0.0
• >lmM| Ull II lillnll
1.0 2.0 3.0
ill 4.0
llll'J '•
-I
5.0
Serviceability Index
IV ~ IV
R = <$> = 4>Uo) (7)
(o- Pf +
°pt>
where § = standard normal distribution; upf, upt = mean values of pf, p,;
apf, dpt = standard deviations of pf, p,; and z0 = standard normal deviate.
Reliability defined in Eq. 7 can be calculated using the following formu-
lation:
300
1 p dz
exp (8)
Performance/Expected Life
Whereas the reliability of a pavement (toward the end of its design life)
is a basic measure of performance, the expected life (average life through-
out which the serviceability does not fall below a minimum level) turns out
to be a better parameter indicating performance. The term expected is to
be understood as in the theory of probability and has the meaning of
"average in a large population" with all relevant circumstances un-
changed. Performance is a measure of the accumulated service provided
by the pavement. In probabilistic terms, it may be interpreted as the
expected life of acceptable service. The minimum level of acceptability
may be defined by the designer; the AASHTO Interim Guide design has
adopted a minimum serviceability index of 2.5, for example. By the
definition of Eq. 3, reliability less than one, say R, means that 1 — R
fraction of the sections has fallen short of providing acceptable minimum
service. Another interpretation of reliability, employing the concepts of
301
Eq. 1, may be offered here. Reliability i?, in the z'fh year signifies that the
life expectancy during that same one year period is Rf (note i?, S 1).
Adopting this point of view and considering the reliability of pavement in
one year increments, expected service life or performance of pavement,
designed for / years, is
E(L) = 2 R, (9)
/=i
CONCLUDING REMARKS
ACKNOWLEDGMENT
The rigid pavement (JPC) model (Eq. 10) was derived from 76 pavement
sections located in 11 states (George et al. 1988). Mechanistic variables
were introduced by using Westergaard's edge stress model directly in the
regression analysis. Climatic variables were included by using them
directly in the regression analysis.
Mp
Pt = Pi ~ 3 exp - 1^1 (10)
where p,, /?,• = terminal, initial serviceability index; p = 19.7 F + 9.4 Dow
- 48.9; N = ESASL + 0.00256 Age * FI; and (J = 1/(2.42 Fm - 0.004 FI
- 0.009 AP) in which Dow = 1 (dowel used), 0 (no dowel); ESAL = 18-kip
equivalent single axle load; FI = freezing index, degree days; F = stress
ratio, fthe; with/, = modulus of rupture of concrete; and ae = Wester-
gaard's edge stress.
303
304
306