You are on page 1of 4

2006 IEEE Congress on Evolutionary Computation

Sheraton Vancouver Wall Centre Hotel, Vancouver, BC, Canada


July 16-21, 2006

Optimization of pavement design using a genetic algorithm


Andy Pryke, Harry Evdorides and Rawya Abu Ermaileh

Abstract— The design of road pavements is an engineering ultimate aim of flexible pavement design [1], and there are
problem with a number of required design inputs. Although typically several designs which satisfy the requirements,
many analytical methods are in use, the existing methods do not each with differing construction costs.
explicitly include optimization processes whereby the design of Several design methods have been developed which help
pavement may be linked to criteria and associated objectives
different from the conventional engineering-driven ones. This
in the choice of the optimum thickness and material of the
paper presents a prototype system that demonstrates how an different layers for a given design specification. Pavement
analytical pavement design procedure can be optimized using a design methods are either empirical or analytical. Generally
genetic algorithm. The approach considered is flexible as it can the former require the designer to refer to design tables,
be used to optimize any computerized pavement design method. charts and formulas, and the latter use a computer model in
In addition, the results obtained from the system are conjunction with performance relationships to define a
comparable with those derived from a well known method. It is
felt therefore that this exemplar prototype system may be
suitable design. However, designers may stop after a few
further developed to cater for any type of road pavements and iterations, before an optimum design is reached as the
to include more complex optimization procedures. procedures are not fully automated. In addition the
optimization process is limited to the criteria of the design
I. INTRODUCTION method used and may not include other external criteria such

R oad pavements are complicated structures which are as cost. Here we present a simple system which automates
subjected to a variety of traffic loading and the search for a multi-objective optimum pavement design,
environmental conditions. Pavement design systems aim to replacing the need for manual search.
cope with this complication in an effective and efficient
way. The design procedure used should enable the selection III. THE PROPOSED DESIGN PROCESS
of a pavement design that provides adequate performance as Automated design optimization requires the following:
well as adequate distribution of life cycle cost [2]. As there • An objective function to be optimized .
are many design options, searching through these to identify • A set of design variables which affect the value of the
the optimum solution is often time consuming and objective function.
expensive. • A set of design constraints.

The work described in this paper aims to investigate how In this case, we decided that:
the optimization of flexible pavement design may be • The objective function should aim to minimize the
facilitated using evolutionary algorithms. In particular, the construction costs of the pavement.
process seeks to optimize the selection of thickness and • The design variables would be the layer thicknesses
material for the pavement layers in addition to finding the and material choices.
most economical pavement structure for the input traffic • The design constraints would be the design
data. parameters that are associated with the ability of the
pavement to withstand a given level of traffic,
II. PAVEMENT STRUCTURAL DESIGN maximum and minimum thicknesses for particular
The design of flexible pavements involves a study of soils materials.
and paving materials, their behavior under load and the
design of the pavement structure to carry that load under all The output of the system would be the optimum design
climatic conditions [9]. The philosophy of the analytical found, consisting of: the number of layers; the combination
approach to pavement design is that the structure may be of materials; the thickness for each layer; and the total cost
treated in the same way as other civil engineering structures, for the pavement.
the procedure for which may be summarized as follows [4]:
A. Design Constraints - Analyzing Pavement
specifying loading; estimate size of components and
Performance
determine materials; structural analysis to estimate
performance. This procedure is iterated with changes to The design of pavement structures is carried out with
materials or geometry until a satisfactory design is achieved. regard to the ability of the pavement to withstand traffic and
Optimizing the thickness of pavement layers is the environmental loading during a predefined period of time
(design life) which is usually 20 years. This is expressed in
terms of the number of standard axle loads to failure.

0-7803-9487-9/06/$20.00/©2006 IEEE 1095


Conventionally, two forms of failure are considered, fatigue the required constraints.
cracking and rutting. Several computer models capable of
computing stresses and strains associated with these two 3) Design Constraints and Control Variables
defects for any given combination of loading, materials, and The optimization is subject to the following design
layer thicknesses have been developed and widely used. constraints:
Models can be based on elastic, visco-elastic theories or on t min ≤ ti ≤ t max and ti ≤ t drainage (for drainage layers) and
finite element analysis techniques [7]. However in this study
the Method of Equivalent Thickness (MET) was used to n ≤ ( N f and n ≤ N r )
calculate pavement performance because it is simple, fast to where [tmin, tmax] is the practical range of thicknesses of
calculate and most importantly it gives reasonably accurate various pavement layers; tdrainage is the required thickness
results. It is worth noting however that MET can only model determined by the drainage criteria; n is the required design
pavement structures in which the stiffness of materials life in load repetitions; Nf is the number of load repetitions
decreases with depth and that its accuracy decreases with to fatigue failure; Nr is the number of load repetitions to
increasing the number of layers. Only a single operating rutting failure. Nf and Nr are calculated using MET and an
temperature is considered, and the effect of temperature on appropriate pavement performance model.
the stiffness of asphalt layers has been incorporated into the
material properties. 4) Representation of Design Parameters
Solutions are encoded as binary strings, with each numerical
B. Optimization Process
value taking up 5 bits, giving 32 possible values. Future
The process aims to find the optimum combination of work will include real value representations, direct
pavement layers and available materials which can carry the representation of variable numbers of layers, and choice of
expected traffic safely during the design period. An optimal materials.
design is defined as the one that most economically meets
these performance requirements. 5) Evolutionary Process
The system as a whole allows analyses of pavement Basic Algorithm
structures with between three and to six layers. For each 1) The initial population of solutions is generated at
layer up to three alternative materials can be tried. The random;
system uses Genetic Algorithm (GA) optimization to find a 2) Designs that violate constraints are replaced
solution for each number of layers and each possible 3) A fitness value is calculated for each individual in the
combination of materials. These solutions are compared and population.
the most optimal is identified. Future work could integrate 4) Roulette-wheel selection is used to select parents
these choices into the genetic algorithm optimization. 5) Single point crossover and mutation are applied with
GAs are a popular method for a variety of complex specified probabilities.
optimization problems. They require little knowledge about Constraint handling
the problem being solved, are easy to implement, robust, and In addition to the design criteria, the system allows users
inherently parallel [3]. to specify constraints on the materials to be used and their
possible combinations along with the maximum and
1) Objective Function and Design Variables minimum layer thickness. These reflect the practical
In this paper, only the initial construction costs of building constraints for the local conditions. During execution,
the pavement structure is considered, though predicted solutions that violate the constraints are discarded. This is
maintenance costs could be incorporated in future work not problematic as there are many feasible solutions
Mathematically, this is expressed as the objective function generated. Although it is possible to incorporate a fitness
formulated by the equation below penalty method [5], highly constrained problems are not
C c (Ta ) = ∑ t i × u i considered in the present study, and the system appears to
perform well with this simple method.
where Cc(Ta) is the initial construction cost of original
pavement structure in £/m2. ui is the unit cost of the 6) Parameterization of the GA
pavement material in £/m3; ti is the thickness of various Population size
pavement layers. The population size was set at 200. The analysis
performed indicated that larger population sizes did not lead
2) Design Variables to much improvement in results.
The design variables are the thickness of pavement layers
(ti ), choice of pavement materials (which determines unit
cost, ui ), and the number of layers within the pavement
structure. These factors can be varied to acquire an optimum
design which minimizes the construction cost and satisfies

1096
TABLE 1
INPUT DATA OBTAINED FROM SHELL MANUAL [8]

Genetic operators Parameter Value


Basic one-point crossover and a simple mutation operator Cumulative number of ESA 60, 35, 18, 6
are used in the system and their probabilities have been set or 1.2 Msa
according by trial and error. The sub-grade Strength 50MPa
Stopping criteria Poisson’s ratio of sub-grade 0.35
Modulus of unbound granular layer 400MPa
The process is stopped after 200 generations. This number
Poisson’s ratio of unbound granular layer 0.35
was determined by running the system on various test Unit cost of granular material 72£/m3
problems. The value was chosen to strike a balance between Modulus of asphalt layer 5000
allowing enough generations to reach asymptotic Poisson’s ratio of asphalt layer 0.35
performance and a having a reasonable amount of Unit cost of asphalt material 20£/m3
computational time for each run.

IV. TESTING AND COMPARISON


TABLE 2
COMPARISON WITH SHELL DESIGN METHOD DESIGNS
A. Comparison with Other Methods
To validate the effectiveness and the efficiency of the
Traffic Load
approach methodology, the results from the system were 60 35 18 6 1.2
(Msa)
compared with those from existing design methods using
Shell Asphalt thickness
data found in the literature. The design data used in this (Sub-base 270 250 230 190 140
(mm)
system, for example the traffic data, sub-grade condition, = 200mm Cost
paving materials, their properties, drainage data and cost (£/m2)
23.5 22 20.6 17.7 14.1
have been selected for a typical single carriageway with two GA Asphalt thickness
lanes road. A full specification is shown in table 1. (Sub-base 270 240 210 180 130
(mm)
= 200mm)
Cost
1) Shell Design Method 23.5 21.3 19.1 16.8 13.5
(£/m2)
The Shell Design Method [8] specifies a standard GA Sub-base
thickness for the sub-base layer of 200mm. For comparison (Sub-base thickness 450 420 450 320 250
purposes, the GA based system was evaluated twice, once optimized) (mm)
with a fixed sub-base thickness of 200mm and once with the Asphalt thickness
160 140 120 120 100
thickness of this layer allowed to vary. The thicknesses of (mm)
unbound sub-base layers and asphalt layers shown in table 2 Cost
20.5 18.5 18.2 14.8 12.2
were obtained from charts in the Shell design manual. The (£/m2)
analysis is carried out using both the prototype system and
the shell design criteria and the results can be shown in table
2 and figure 1
The results show that the GA-based system gives a
reduction in the pavement thicknesses and hence in the total
cost. However, care must be taken in this comparison since
different analysis models have been employed in the two
methods.

Fig. 1. Cost Comparisons across design methods

1097
[8] Shell International Petroleum Company Limited (1978), Shell
Pavement Design Manual, London
V. CONCLUSIONS AND FURTHER WORK [9] Yoder, E.J. and Witczak, M.W., (1975), Principles of Pavement
The pavement optimization problem developed in this Design, Second Edition, John Wiley & Sons, Inc.
study has a simple analytical model which could be
approached using a hill climbing method . Nevertheless, this
methodology can be considered as a platform to build on for
more complex models as well as other objective functions or
multi-objective functions which can be used in the pavement
management systems.
As total life cycle cost could be incorporated in the cost
prediction function. The system has been developed for
flexible pavements, it could be modified for rigid pavement
design. Finite element method and layered elastic theory
could also be investigated. These models have a longer
running time but are suitable for a wide range of problems.
The GA based optimization could be improved by the use
of a real number, rather than binary, representation and by
incorporating the choice of materials into the chromosome
representation.
Finally, it should be noted that in spite of these
shortcomings, a GA-based methodology provides an
effective and practical tool in solving pavement design
problems.

ACKNOWLEDGMENT
The concept of pavement design optimization using
genetic algorithms was effected by the Transport
Collaborative Research Network of the University of
Birmingham and developed by R. A. Ermaileh[6]. The
contribution of Mr. Jinping Li, visiting fellow at the
University of Birmingham, to the development of the model
is acknowledged.

REFERENCES
[1] Attoh-Okine NO, Roddis KWM (1994) Pavement thickness variability
and its effect on determination of moduli and remaining life.
Transportation research record, no. 1449. National Research Council,
39–45.
[2] Azmy, O. E., Sharaf, E. A., and Lotfi, H. L., (1989), Combined life
cycle cost and performance approach for selection of optimal flexible
pavement strategies, Transportation Research Record 1216, National
Research Council, Washington D.C., pp. 18-28.
[3] Benjamin, B., Lucken, V., Sotelo, A., (2005), Multi-objective pump
scheduling optimization using evolutionary strategies, Advances in
Engineering Software, v 36, n 1, January, Evolutionary Optimization
of Engineering Problems, p 39-47
[4] Brown, S. F. (1984), Computer Programmes for the Analytical Design
of Asphalt Pavements, Highways and Transportation, v 31, n 8-9,
Aug-Sep, 1984, p 18-23, 26-27.
[5] Carlos A. Coello Coello (2002), Theoretical and Numerical Constraint
Handling Techniques used with Evolutionary Algorithms: A Survey
of the State of the Art. Computer Methods in Applied Mechanics and
Engineering, 191(11-12):1245-- 1287, January 2002.
[6] Ermaileh R. A., (2005), Pavement Design Optimization Using
Evolutionary Algorithms, M.Sc Thesis, University of Birmingham,
UK
[7] Mamlouk, Michael S.; Zaniewski, John P., and He, Wei. (2000),
Analysis and Design Optimization of Flexible Pavement, Journal of
Transportation Engineering, v 126, n 2, Mar, 2000, p 161-167.

1098

You might also like