Subido por FRANCISCO GRANDEZ VENTURA

Airport pavement management systems an a

Anuncio
Transpn Res.-A, Vol. 32, No. 3, pp. 197±214, 1998
# 1998 Elsevier Science Ltd
All rights reserved. Printed in Great Britain
0965-8564/98 $19.00+0.00
Pergamon
PII: S0965-8564(97)00008-6
AIRPORT PAVEMENT MANAGEMENT SYSTEMS: AN APPRAISAL OF
EXISTING METHODOLOGIES
MICHEL GENDREAU and PATRICK SORIANO
Centre de recherche sur les transports and DeÂpartement d'informatique et de recherche opeÂrationnelle, Universite de
MontreÂal, C.P. 6128, succursale Centre-ville, MontreÂal, QueÂbec, Canada H3C 3J7
(Received 24 May 1994; in revised form 16 February 1997)
AbstractÐAirport pavement management systems (APMS) are computer-based decision support systems
that can be used by the agencies running airports to determine cost-e€ective maintenance and rehabilitation
strategies to preserve the various pavement structures (runways, taxiways, etc.) which are a critical component of these facilities. In this paper, we describe the main elements of APMS and review existing systems.
# 1998 Elsevier Science Ltd. All rights reserved
1. INTRODUCTION
Air®elds and highway networks, comprising both streets and roads, constitute an enormous
investment of public funds. In addition, they are the backbone of one of the most important economic activities in modern industrialized societies: the transportation of goods and persons. Preserving this investment and maintaining it in an adequate condition for it to perform its role is
clearly a worthy objective. However, time and usage have taken their toll on these pavement
structures and more and more of them are approaching or have reached the end of their design
life. This deterioration process has progressively brought the agencies responsible for pavement
management to shift their emphasis from construction of new pavements to maintenance and
rehabilitation (M&R) of existing ones. But the management of an aging pavement infrastructure is
a dicult task, given the extreme complexity of pavement behaviour. This task is further complicated by the fact that pavements are being exposed to continuously increasing weights and
volumes of trac, which is accelerating their deterioration, and also because of the growing budgetary constraints under which most agencies are now forced to operate.
This situation has generated, over the last 20 years, a steadily rising interest and research thrust
into pavement management methods, which has brought about the development by many agencies
of pavement management systems (PMS). These systems are designed to provide a structured and
comprehensive approach to pavement management. Their role is to assist decision-makers in
®nding strategies for providing and maintaining pavements in a serviceable condition over a given
period of time in the most cost-e€ective way possible. As stated in Hudson et al. (1992), the
function of a PMS is to improve the eciency of decision-making, expand its scope, provide
feedback on the consequences of decisions, and insure the consistency of decisions made at different management levels within the same organization. However, as evidenced by the published
literature on the subject, PMSs can take on many di€erent forms, depending on the particular
organization and preferences of the agency within which they are implemented. Nevertheless, they
all share a set of common features or functions, essential to their operation. These essential features are: network inventory, pavement condition evaluation, pavement performance prediction,
and management planning methods. In this paper we will describe these elements and review the
di€erent ways in which they are implemented in practice, highlighting as much as possible their
strengths and weaknesses.
The objective pursued here is to provide a synthesis and an appraisal of the di€erent methodologies used in airport pavement management systems (APMSs). However, the existing literature on APMSs reviewed in the course of this study have revealed that very few di€erent systems
have been the subject of published scienti®c papers. In addition, all of these systems consist of
197
198
M. Gendreau and P. Soriano
limited enhancements of, or variations on, one particular APMS approach, namely that of the
PAVER system (Shahin and Kohn, 1982a), and hence use essentially the same methodologies.
Although we are speci®cally interested in APMSs, most existing PMSs have been developed for
highway networks. However, the concepts and methods used in these are often valid in the airport
context as well. We will therefore also consider in this study methods drawn from the highway
setting that are applicable in APMSs.
The paper is organized as follows. The network inventory function is discussed in Section 2. In
Section 3, we review the di€erent methods used in existing PMSs to evaluate and rate the condition of pavements. Section 4 focuses on the approaches used for predicting the evolution of pavement condition over time, while Section 5 addresses the planning aspect. Other interesting features
found in some of the more complete systems are described in Section 6. Finally, Section 7 concludes the paper.
2. NETWORK INVENTORY
The ®rst and most basic function of a PMS is to provide a complete and structured inventory of
the pavement network to be managed. This is accomplished by dividing the network into relatively
small units called sections. These sections represent the minimum fraction of the network for
which major M&R decisions are made. Each section is de®ned in order to exhibit consistent
characteristics, such as pavement structure, construction history, functional classi®cation (e.g.
primary runway, apron, secondary taxiway, ...), trac volumes and mixes, and condition (Shahin,
1980; Butt, 1991). In several cases, runways and taxiways are also divided laterally into parallel
sections to account for the channelization of trac, i.e. lateral distribution of trac across runways and taxiways according to a bell-shaped distribution (Herrin et al., 1986; Lary and Soule,
1991; Shahin and Kohn, 1982b). The network inventory also includes the identi®cation number
and area of each section. Finally, it is suggested in Shahin (1983) that the consistency of the sections be veri®ed periodically by examining their condition (surface and de¯ection response) for any
systematic change and further divided if that is the case. The PAVER system provides some
guidelines for this with respect to observed surface condition variations (Shahin et al., 1979).
As can be seen, the implementation of this PMS element does not present any methodological
diculty in itself. The main question that it raises lies in the manner and extent to which all the
necessary data are to be combined and integrated. The obstacle here is the very vide variety of
information sources: drawings, tables, charts, maps, texts, etc. (Schwartz et al., l991; Rada et al.,
1992). This can pose many problems depending on the level of ¯exibility and accessibility one
wants to have. In general, all this information is integrated into a computerized database whose
sophistication can vary a great deal. In the most user-friendly systems this database is integrated
within a graphic and interactive environment that allows easy and quick data access or update,
as well as graphical display of information (Schwartz et al., 1991; Rada et al., 1992; ERES Consultants, 1993; PCS/Law Engineering, 1993). We will comment further on these aspects in
Section 6.
3. PAVEMENT CONDITION EVALUATION
To complete the inventory of the network, the managing agency needs to perform an evaluation
of the present condition of the di€erent pavement sections under its responsibility. This evaluation
will then serve as one of the main inputs in the decision process that will determine the M&R
activities to be carried out. It is therefore a crucial element of any PMS. Without accurate data,
PMS outputs would not be very reliable. However, as will be seen in this section, the condition of a
pavement is not a trivial notion: there exist many di€erent de®nitions of pavement condition and,
thus, many ways of evaluating it.
3.1. Structural and functional performance
Originally the design procedures and consequently the evaluation of the condition of airport
pavement structures were centered on structural considerations, such as limiting stresses, strains
and de¯ections in one or more critical layers (Monismith, 1977, 1978). However, it has been
recognized for some time now that pavement evaluation should also address the functional
Airport pavement management systems
199
performance of the structure (Monismith, 1978; Witczak, 1978), i.e. how well it ful®lls its role of
providing aircrafts (or vehicles in the case of roadways) with a safe and comfortable ride. This
aspect of pavement performance is very important in the context of air®elds and particularly so
when considering runways. Indeed, contrary to most other pavements, where one always has the
alternative of reducing its speed when confronted with a rough pavement, this option is not
available for runway operations since a threshold velocity must be attained for both takeo€ and
landing. Therefore, if a runway surface becomes too uneven to allow safe (or comfortable)
operations, then that pavement can no longer be considered adequate, regardless of its structural
capacity.
Although structural and functional performance are related, no well-de®ned relation between
them is known to date (Monismith, 1978; Zaniewski, 1991). Hence, any rational pavement evaluation procedure should consider both of these aspects of pavement performance in order to be
e€ective and useful.
3.2. Structural evaluation
Measuring the structural (or bearing) capacity of a pavement section is a very technical subject
that has been widely studied and documented in the specialized literature, but whose detailed discussion lies beyond the scope of this paper. For an overview of the state-of-the-art in this ®eld and
an extensive list of references, the reader is referred to Zaniewski (1991). We will simply note that
this evaluation consists of determining the physical characteristics of the materials composing the
pavement structure (California Bearing Ratio, modulus of elasticity, subgrade reaction modulus,
etc.) and then analyzing the e€ects of loadings on the structure to assess its deformation response.
This analysis is performed using established engineering models developed for the di€erent pavement types. These models are either based on empirical relationships, such as the California
Bearing Ratio (CBR) method, or on more elaborate mechanistic-empirical methods, generally
using elastic layer theory or thin plate theory for their analysis of pavement response to loadings.
The actual consensus in airport pavement management is that the latter should be preferred
(Shahin, 1983; Zaniewski, 1991). Other mechanistic methods based on di€erent material behaviour
theory or analysis techniques have been proposed, but have yet to gain widespread acceptance.
The reader is referred to Monismith (1978) and Zaniewski (1991) for a thorough description and
discussion of this topic.
Evaluating the structural capacity of pavements provides important information that is then
used in several ways (Witczak, 1978):
(a) To determine the allowable load that can use a speci®c pavement for a given predetermined
life. This, in turn, can serve to control the access of speci®c aircraft types to some pavements
in light of the actual strength of the pavement, and also to establish landing fees related to
the amount of structural life used by each aircraft type, as is done with the ACN/PCN
concepts described in Antunes and Pinelo (1990) and Alexander and Hall (1991).
(b) To estimate the remaining life of a pavement at a given time and aircraft-trac history as
will be discussed in the next section.
(c) To assess the strength of the existing pavement and eventually determine the future overlay
requirements when considering rehabilitation programs.
The data required to do this evaluation have traditionally come from construction records and
both on-site and laboratory destructive testing (i.e. coring, trenches). However, in a growing
number of cases, the bulk of this information is now gathered via nondestructive testing (NDT)
methods, among which the Falling Weight De¯ectometer (FWD) is probably the most widespread
across North America. Other NDT methods that can be mentioned are the Heavy Weight
De¯ectometer (HWD), which is specially useful in the airport context since it generates loadings
much more comparable with the actual loadings induced by aircrafts, ground penetrating radar
and infrared thermal photography which are particularly useful to locate voids beneath the pavement surface (see Hudson and Uddin, 1987 for further details). These NDTs, in addition to the
obvious advantage of leaving the pavement undamaged, can also be performed much more rapidly
and more economically than traditional destructive testing. This, in turn, can enable agencies to
conduct more extensive and/or frequent surveys of their pavements' structural capacity, thus providing them with a better basis for decision-making.
200
M. Gendreau and P. Soriano
Finally, computer programs that perform quite sophisticated pavement response analyses,
almost directly from FWD or HWD test readings, are now well established and often easily
accessible. Hence, one could be drawn into believing that the structural evaluation of pavements is
nowadays exempt of diculty. This is unfortunately not the case, however. If the data collection
and analysis problems have in a way been settled, the interpretation of the results from these
analyses is far from trivial and demands extensive experience and knowledge (Tessier, 1990).
Indeed, as pointed out in the conclusion of Zaniewski (1991), given the discrepancies that exist
between the true behaviour of the materials composing the pavement and the assumptions
underlying the di€erent analytical models used to describe it, the results of these de¯ection analyses are bound to be subjected to a fair amount of variance and should, therefore, be interpreted
with caution.
3.3. Functional evaluation
As stated in Zaniewski (1991), one of the most dicult aspects of air®eld pavement analysis is
the de®nition of functional pavement failure. Indeed, excess stress of the pavement structure
results in fracture of the materials. However, cracking alone does not signify a failure of the
pavement surface since aircrafts can traverse distressed pavements. In the management of highway
networks, pavement failure is usually de®ned in terms of the functional characteristics of the
pavement surface, primarily with respect to roughness (ride quality). No comparable de®nition of
failure exists in the case of airport pavements. In this management context, it is the accumulated
e€ect of di€erent types of distress and their impact on several functional characteristics that is of
major concern. It is now generally admitted that rational evaluation procedures of pavement
functional performance should be based on the analysis of roughness, skid resistance, the potential
for foreign object damage (FOD) to aircrafts, as well as surface distresses (Shahin, 1980, 1982).
The evaluation of these characteristics can be made through direct measuring, visual condition
surveys, or a combination of both. In most cases, these measurements or surveys are then expressed in the form of a quality index. These indices are either objective (i.e. based on a direct quanti®ed measurement) or subjective (i.e. based on judgment). They can correspond to a single or to a
combination of some or all of the pavement characteristics that are evaluated by the agency; they
are then respectively referred to as individual, composite or overall indices (Tessier, 1990; Baladi et
al., 1992). These indices de®ne the state of deterioration of the pavement at a speci®c time.
One important point that must be noted about these indices is that there exists very little standardization, particularly among highway pavement management agencies, with respect to their
de®nition, range of values, etc. This stems ®rst from the fact that several technologies are often
available to measure a given characteristic. For instance, pavement roughness is typically measured with either a pro®lometer or a response-type measuring system (Shahin, 1980; Tessier, 1990;
Baladi et al., 1992), for both of which several models exist, each having their own speci®c characteristics (see Tessier, 1990 for further details). But more signi®cantly, this wide range of indices
illustrates the dicult and rather subjective task of de®ning a relationship between the di€erent
surface characteristics and the functional condition of pavements. These numerous indices re¯ect
the varied perceptions of this relation that exist among pavement management organizations. A
list of several of these methods for highway agencies can be found in Baladi et al. (1992). Other
indices of functional performance are described in Kher and Cook (1985), Majidzadeh et al.
(1992), Mosheni et al. (1992) and Ullidtz (1985).
For airports, however, the situation is quite di€erent. In APMSs, the performance evaluation
procedure developed in the late 1970s by the U.S. Army Corps of Engineers (USACE) and on
which the PAVER system (Shahin et al., 1979; Shahin, 1982, 1992; Shahin and Kohn, 1982a;
Shahin and James, 1984; Hall et al., 1992) is based, is now extremely widespread and constitutes a
sort of de facto norm in the ®eld. Though other methods probably exist, it is the only one for
which published literature was found and serves as the basis for the functional evaluation procedures of all the APMSs reviewed in this study. This method, which was adopted by the Federal
Aviation Administration (FAA) for use on civilian air®elds in 1980 (Federal Aviation Administration, 1980), is centered around an overall surface condition index, the pavement condition index
or PCI. The PCI is a numerical indicator of pavement condition, whose values range from 0 to 100
(excellent condition). It uses weighted deduct values that are functions of the type, severity, and
extent of visible distresses, to combine data on individual distress types into a single condition
Airport pavement management systems
201
value. These data are collected through visual surveys with some direct measurements to evaluate
the severity of some distress types, such as rut depth. It is, therefore, a semi-objective index. To
expedite the survey process, each section is subdivided into sample units and only a random portion of these is evaluated. The PCI value of the section is then computed as the average PCI of the
inspected sample units of that section.
As can be seen, the PCI is essentially a surface distress index and, as such, does not constitute a
comprehensive functional performance indicator. However, surface distresses have a signi®cant
in¯uence on the functional condition of pavements and the PCI can therefore be used as a means
of assessing this condition, even though it is not a direct measure of it. Note that, since surface
distresses are also viewed and used as indicators of the structural condition of pavements, the PCI
provides, in addition, a standard method for rating the structural integrity of pavement sections
(Shahin, 1980). In fact, this was one of the main objectives that motivated the development of the
PCI methodology.
The PAVER evaluation procedure (Shahin et al., 1979) and those derived from it (e.g. ERES
Consultants, 1993 and PCS/Law Engineering, 1993) place a major emphasis on the PCI and the
distress data collected during the visual survey, but they do also address the other important
aspects of pavement performance that are roughness and skid resistance by speci®cally including
in the evaluation process the results of direct measurings of these characteristics. Though regularly
monitored by managing agencies in order to meet the standards set by the International Civil
Aviation Organization (ICAO), these measurings are not often integrated into the pavement
management process. Furthermore, these evaluation processes (in particular the PAVER system)
generally incorporate several other measures, such as:
(a) variations of the PCI within a section, providing also some statistically-based guidelines to
determine if these variations can be considered localized or systematic, in which case further
division of the section in more homogeneous components is required;
(b) rate of deterioration, which enables the identi®cation of rapidly degrading sections;
(c) distress cause distribution (as related to load, climate or other factors) to identify the primary cause of distress.
Since, in addition to all of this, the results of the structural evaluation are included in a single
framework, they therefore constitute a rather extensive and complete evaluation process that
should provide managing agencies with a good picture of the overall condition of their pavement
networks.
One particular strength of these approaches, and of the PCI in particular, is that they force the
compilation of a wide array of data on the individual distresses in order to compute the overall
index. This information can then be kept and accessed, if needed, later when determining precise
M&R actions to be taken. However, there are some drawbacks and weaknesses in the PCI
method, one of them being the somewhat questionable repeatability of the visual surveys, which is
caused by the frequently subjective nature of the di€erentiation between distress severities. This
can be corrected through more extensive and rigorous inspection guidelines, as reported in the
Indiana airports experience (Eckrose and Reynolds, 1992) or with the use of automated data collection devices (Cation and Schwandt, 1991; Lee, 1992).
4. PERFORMANCE PREDICTION MODELS
Predicting future pavement performance and condition is understandably of signi®cant concern
to pavement engineers. Indeed, with the now predominant emphasis on M&R activities, PMSs
have become more and more dependent on the estimation of the future performance of existing
pavements. The necessity of reliable performance prediction models is therefore greater now than
ever before.
In modern PMSs, pavement prediction models are absolutely essential elements that in¯uence
many critical management decisions. At the project level, they are used to design pavements, to
perform life-cycle cost analyses, to select optimal design with least cost and, in tradeo€ analyses in
which the annualized costs of new construction, maintenance, rehabilitation, and user costs are
summed for a speci®c pavement design, to determine the best time and pavement condition in
which to perform each (Lytton, 1987). At the network level, they serve ®rst and foremost in the
202
M. Gendreau and P. Soriano
selection of optimal M&R strategies, in short- and long-term budget optimization, and in determining equitable fees (e.g. landing fees related to predicted damage; see Alexander and Hall, 1991)
as well as inspection schedules to monitor pavement deterioration.
There are two basic types of pavement performance prediction models: they can be either
deterministic or probabilistic (Lytton, 1987; Butt, 1991). The deterministic models predict a single
number for the remaining life of a pavement or its level of distress or whatever measure of its
condition being predicted. It thus considers the evolution of pavement deterioration over time as a
completely predictable process. On the other hand, probabilistic models predict a distribution of
such events, thus describing the di€erent possible future conditions as the outcomes of a stochastic
process. Here, the deterioration process is considered to be somewhat uncertain and therefore not
amenable to exact prediction.
4.1. Deterministic models
Deterministic models are the most widely used of the two types of prediction models. Among
them we can distinguish between structural and functional performance prediction models,
depending on the type of prediction carried out.
4.1.1. Structural performance models. These models are the natural extension of the pavement
behaviour models mentioned in Section 3.2 and they are used to predict individual pavement distresses of all kinds. As was stated then, they may be essentially empirical or mechanistic-empirical,
that is, based on a mechanistic model of the materials response calibrated with observed ®eld data
(hence the term empirical) and on damage accumulation models. These types of models essentially
relate the materials characteristics of the pavement structure and the loads applied to it, in order to
determine the number of cycles of load applications before failure occurs. The failure criterion
used here is de®ned according to the type of distress being studied and can be one of several criteria, such as ®rst apparition of fatigue cracking, a predetermined rut depth, or other. Combining
these models with the most recent materials characteristics measures (as determined from de¯ection measures, for instance) and expected future trac levels and mixes then enables managing
agencies to predict the structural remaining life of pavement structures. This, in turn, serves in the
planning of future M&R actions. Prediction models of this type have been developed for di€erent
pavement types by several organizations like the Portland Cement Association, the Asphalt Institute, the USACE and the Shell International Petroleum company (see Monismith, 1978 and
Zaniewski, 1991), and are extensively used by pavement engineers. Among the APMSs reviewed,
both PAVER (Shahin, 1983) and IAPMS (Rada et al., 1992) explicitly mention that such structural prediction models are integrated into their systems.
4.1.2. Functional performance models. Most of the models in this category are found in highway
PMSs and are used to predict the present serviceability index (PSI), pavement surface friction (skid
resistance) or wet-weather safety index (hydroplaning potential). Like the structural models, they
are either empirical or mechanistic in nature. The COPES (Concrete Pavement Evaluation System)
and NCR models (Rooke, 1985) fall into this category, as do the models used in the PARS system
of the province of Ontario (Kher and Cook, 1985) and in DMS, the PMS of Danemark (Ullidtz,
1985).
In the APMS context, the only type of functional performance prediction models used are PCI
prediction models. Most of these have been developed for the PAVER system but can and are
being used within other APMSs, as is the case with IAPMS, ERES DSS, and AIRPAV. However,
contrary to their highway counterparts, these models are essentially empirical and not based on
mechanistic models. They do not try to base their prediction on an analysis of the pavement
degradation process that would attempt to relate loading and climatic history to surface conditions, but merely to ®nd a relation between the PCI and other accessible data characterizing the
pavement structure. Hence, the prediction process that de®nes this type of model rests on previously observed data and must therefore be determined or, at least, calibrated through some sort
of statistical analysis procedure.
In its simplest form, as was the case in the early versions of the PAVER APMS (Shahin et al.,
1979; Shahin, 1982), the prediction is performed by a straight-line extrapolation using only two
previous observations of the PCI. The future values of the index are determined by the straight line
Airport pavement management systems
203
passing through these two points on a PCI vs time chart. No other variables are considered in the
prediction process. This method is, however, too simplistic and too inaccurate to still be considered today as something other than a temporary solution while other more adequate techniques
are being developed.
The most common statistical analysis technique for prediction model building is certainly multiple regression analysis, which has been used in many contexts to de®ne and calibrate such models. In essence, models built with this approach relate the future PCI value to a series of
explanatory or predictive variables, such as the age of the pavement structure, the time since its
last overlay, de¯ection testing information and trac measures among others, by way of a mathematical expression often termed predictive equation. During the development of the PAVER
system such equations or models were de®ned by the USACE for both rigid and ¯exible pavement
by analyzing an extensive set of data taken from several U.S. Air Force bases (Shahin and Becker,
1984). Their predictive performance was found to be quite good in the higher values of the PCI
scale, but dropped signi®cantly for pavement features with PCI less than 65 or 50 (for rigid or
¯exible pavements, respectively) (Shahin and Becker, 1984; Shahin and James, 1984). In these
studies and in similar ones carried out with the Ontario highway PMS (Kher and Cook, 1985), it
was also found that universal models developed with data taken from several geographical regions
were generally outperformed by local ones. By using only local data, the resulting model will
implicitly incorporate speci®c climatic, soil, construction, and materials variables that cannot be
accounted for in the universal models, and, thus, provide better predictions. These local models
o€er the added advantage of being able to be re®ned and updated as new data become available.
Regression analysis is a very powerful tool for prediction model building, but it must be used
with some care. Indeed, the predictive equations must be meaningful with respect to the variables
selected and not just determined in order to best ®t the available data. This is essential if one wants
to obtain a realistic model and be able to have some con®dence in its predictions. One also has to
note that these techniques require relatively large amounts of data in order to provide accurate
models. However, given the extremely complex nature of pavement behaviour, even models
developed with extensive data ®les do not provide very accurate predictions as was just mentioned.
It should also be stated that the prediction capabilities of regression models are de®ned by the
range of the data on which they are developed and do not allow them to extrapolate well beyond
that range. This is one of the reasons why the PAVER models were not accurate for PCI values
less than 65±50: the data that were used to develop them corresponded to pavement features that
were, for the most part, in good condition, and thus the predictive range was limited to the upper
half of the PCI scale.
4.2. Probabilistic models
Probabilistic prediction models include three types of models: survivor curves and simulation
models, that have drawn limited interest, and also Markovian models which, over the last 10
years, have been rapidly gaining recognition and acceptance as a powerful prediction technique in
the highway context and are now being introduced in the airport setting. The motivation behind
these type of models is the observation that the pavement deterioration process and, in particular,
the rate of deterioration are not really deterministic in nature but uncertain. Therefore, a predictive model should portray this process as stochastic, rather than using the erroneous assumption of
deterministic behaviour (Butt, 1991). These probabilistic models, and, in particular, the Markovian approach, provide a rational structure to portray this stochastic aspect of pavement behaviour.
4.2.1. Survivor curves. Survivor curves are used for planning M&R alternatives on pavement
networks (Lytton, 1987). The construction, maintenance, and rehabilitation data recorded by
management agencies are used to develop these curves. They are graphs of probability vs time: the
probability drops o€ with time (or accumulated standard loads) from its initial value of 1.0 down
to 0.0 and expresses the percentage of pavements that remain in service at any particular time
without requiring major maintenance or rehabilitation.
4.2.2. Simulation models. Computer programs based on mathematical models of pavement
response to loads and climate can be developed to simulate the behaviour of pavements, week by
week, over any desired period of time, as described in Ullidtz and Larsen (1983). In this particular
204
M. Gendreau and P. Soriano
model, for instance, the input parameters characterizing the pavement structure, such as surface
elevation of layer interfaces, bitumen content, and so forth, are considered to be stochastic and
vary from point to point in the structure. Hence, the response of the structure is also stochastic.
These programs can be used to predict the future condition of pavements. However, this approach
is based on extensive repetitive calculations that require signi®cant amounts of computer resources. They are, therefore, impractical for prediction in a planning context where numerous simulations would be required.
4.2.3. Markovian models. These types of models are, in a way, the probabilistic counterpart of
the deterministic performance prediction models of Section 4.1.2.: they also predict the functional
condition of pavements, and share as well the fact that they do not aim at analyzing the deterioration process but merely ®nd ways to model it with readily accessible information about the
pavement characteristics.
At the heart of the Markovian prediction model lies the notion of state which is used to represent the condition of a pavement section at any given time. For example, condition states could be
de®ned with respect to the PCI of sections: sections having a PCI in the range 100±91 would be
said to be in state 1, those with PCI between 90 and 81 would be in state 2, and so on. The evolution of pavement deterioration with time is modeled by transitions from one condition state to
another as time progresses. Going back to the example, a section whose PCI value decreases from
95 to 91 and then to 86 over a 3-year span will be modeled as remaining in state 1 for periods
(years) 1 and 2, and then transiting to state 2 in period 3.
However, as was mentioned at the beginning of this section, this deterioration process is considered to be probabilistic in nature: the evolution of pavement condition is governed by probabilities which are associated with the di€erent transitions possible. Each transition probability
represents the chances that a pavement section that is presently in a given condition state will end
up in a speci®c condition state in the next period (i.e. year). These probabilities are generally
expressed in the form of a matrix (a Markovian transition matrix) and one such matrix is de®ned
for each group of pavement sections presenting similar characteristics (e.g. age, construction type,
trac conditions). The argument here is that similar pavement sections are expected to have a
similar deterioration process, even though this process is somewhat uncertain.
The key assumptions underlying Markovian prediction models are:
(a) the number of states de®ned to represent the condition of pavement sections must be ®nite;
(b) the transition probabilities that describe the degradation process depend only on the present
condition state of the pavement section and not on its previous history (note however that
condition state de®nition is quite ¯exible and can include historical information such as the
pavement age, as well as other relevant factors: climatic, environmental, etc.);
(c) the degradation process is stationary, i.e. the probability of making a transition from one
condition state to another is independent of item; for instance, a given transition will have
the same probability of occurring if it occurs in period 5 or in period 1.
This deterioration process is said to be modeled by a ®nite state Markov chain. For further
details on Markov chains and other stochastic processes, the reader is referred to Ross (1970).
In addition, for the Markov process to behave like a realistic deterioration process, one has to
impose that the transition probabilities be null for any transition corresponding to an improvement in pavement condition (without M&R intervention). Hence, the only allowable transitions
are those that either maintain the pavement condition or reduce it, which is the classic pattern of
pavement deterioration. The process is completely determined by the de®nition of its states and of
its transition matrix. These matrices can be constructed for any deterioration process and thus for
any pavement type.
Markov models were ®rst introduced into the ®eld of pavement management by the development of the Arizona highway PMS (Golabi et al., 1982; Way, 1985) which was implemented in
1980. Since then, other highway PMS have also been developed around Markovian prediction
models like those of the state of Kansas (Kulkarni, 1985), Finland (Thompson et al., 1987, 1989),
and the Kingdom of Saudi Arabia (Harper and Majidzadeh, 1992). These types of prediction
models have also been developed to be integrated into already existing systems, such as those
reported in Cook and Kazakov (1987) for the PARS system of the province of Ontario and for
Airport pavement management systems
205
PAVER-based PMSs, as described in Feighan et al. (1987) and Butt (1991). This last reference is,
to our knowledge, the only one that reports the use of this type of prediction models in the airport
management ®eld. The transition probability matrices used in these applications come from either
engineering experience, as is the case in Kulkarni (1985) or, as is generally preferred, from the
analysis of historical pavement condition data. This analysis can be based on a range of di€erent
techniques, such as regression analysis (Golabi et al., 1982; Way, 1985) or non-linear programming (Feighan et al., 1987; Butt, 1991) among others.
There are several advantages in using Markovian prediction models. First of all, when properly
developed these models generally present a better predictive accuracy than their main competitors,
the deterministic prediction curves established through multiple regression analysis. This is evidenced by the results reported in Cook and Kazakov (1987), Feighan et al. (1987), and Butt
(1991). In Cook and Kazakov (1987), comparisons were carried out using four di€erent PCI prediction models: two regression models (deterministic) and two Markovian models. The average
errors in predicted PCI relative to the actual PCI values observed were found to be higher with the
regression models than with the Markovian models, and this error rose much more rapidly as the
lookahead period grew. For instance, when looking 5 years into the future, the error values
observed for regression models Nos 1 and 2 were 8.0 and 7.2 points, respectively, while for the two
Markovian models they stood at 5.5 and 2.9, respectively.
However, it must be stated that in order to build such models, extensive PCI historical data ®les
are generally necessary, these data requirements being even larger than with the regression models
in 4.1.2. In some cases engineering knowledge and experience can be substituted to actual PCI
data, as was the case in the Kansas PMS (Kulkarni, 1985), but real data are preferable. A further
advantage of Markovian models is that projections beyond the limits of the existing data will
continue to behave according to the usual pattern of worsening condition with age, something that
the regression models cannot directly guarantee. Finally, these models can be very easily integrated
into the planning process of most PMSs. In particular they are the natural tools to be used in
conjunction with dynamic programming techniques and Markovian decision processes, which will
be discussed in the next section.
5. PLANNING METHODS
The ®nal component of a PMS is the planning module that enables the managing agency to
determine what M&R actions should be taken, given the current and predicted condition of the
pavement sections within its jurisdiction and the ®nancial resources placed at its disposal. It is with
this element that pavement engineers can establish a program of actions to be performed in the
next planning period and plan future M&R investments, in order to maintain or improve the
condition of their pavement structures. In the present economical context of ever-shrinking
resources, these activities take on added importance since they represent the best tool that pavement engineers have to justify requested funds. Having ¯exible and powerful planning tools is
therefore crucial to the overall e€ectiveness of PMSs.
However, determining the best M&R strategy for a large pavement network over a medium to
long-term planning horizon, while respecting stringent budgetary constraints, is evidently a highly
complex problem. Many di€erent approaches have been proposed to carry out this task. The most
important of these planning methods and their underlying philosophy will be reviewed in this
section.
5.1. Planning levels
It is now well established that pavement management involves two di€erent planning levels that
address di€erent concerns (Kher and Cook, 1985; Lee and Hudson, 1985; Cook and Lytton, 1987;
Gendreau, 1987; Butt, 1991): project and network. At the project level, decisions pertain to when a
pavement section should be rehabilitated and what treatment or action should be performed. At
the network level, one addresses the broader question of where, when, and what M&R actions
should be performed in order to optimize a given criterion, while satisfying some constraints.
Pavement management actually started at the project level in the mid-1960s, with the notion of
pavement design as its focal point (Butt, 1991). The main objective at the time was the determination of the best possible design for each individual project and the network level planning simply
206
M. Gendreau and P. Soriano
consisted in accumulating all the projects for each year of the planning horizon. This approach
was referred to as `from the bottom up' and it worked quite well when sucient funds were
available. However, with increasingly limited budgets and the shift from construction activities to
M&R experienced in the 1970s, more and more projects had to be deferred or even abandoned.
Consequently, it rapidly became apparent that pavement management also had to perform an
important function at the network level in order to adequately assess network condition, M&R
needs and budget requirements. This led to a `from the top down' approach to network level
pavement management that is now prevalent in many of the highway PMSs developed over the
last 10 years, and is now being introduced in APMSs (Butt, 1991). In this approach, the action
planned for each individual section need not be the best possible one for that section, considered
independently of the others (as is the case in the bottom-to-top approach) but it must be the one
which, when combined with all other decisions, provides the best overall network result.
5.2. Project level planning
As we mentioned above, the project level management deals with essentially technical management concerns (such as detailed design decisions) regarding an individual project, or section, considered in isolation from other projects. It requires detailed data on the individual pavement
sections that include loads, environmental and climatic factors, material characteristics, construction and maintenance variables, and costs. The main types of approaches for this planning problem are based on engineering judgment, life-cycle cost analysis, and dynamic programming. In
addition, experimental versions of expert systems have also appeared recently. We will now brie¯y
review these techniques.
5.2.1. Engineering judgment. Most of the older PMSs used (and some still do) some form of
engineering judgment to address the decisions faced at this planning level. This judgment could be
expressed through simple guidelines or through rather complex decision trees or tables requiring
extensive condition data. The methods found in the standard PAVER systems for both air®eld
and highway pavements fall into this category (Shahin et al., 1979; Hall et al., 1992; Lee and
Bowen, 1992). So do the airport management system IAPMS (Rada et al., 1992) and many highway PMSs. The main advantages of this approach is that it is easily implemented, generally without requiring any additional resource from outside the agency, and that the prescribed M&R
alternative addresses the speci®c de®ciencies found in the pavement. The disadvantage, however, is
that the recommended M&R strategy may very well not be the most cost-e€ective one since it
corresponds to a pre-established choice.
5.2.2. Life-cycle cost analysis. This method of selection between alternative M&R strategies for a
given section or project has become the most common for tackling the project planning problem.
It is designed to analyze potential M&R alternatives that are each capable of providing some
required performance and to identify the one that minimizes a given economical criterion evaluated over their expected life. The di€erent alternatives to be considered are identi®ed through
traditional engineering methods. Several di€erent criteria may be used such as present worth,
annualized cost or bene®t±cost ratio. This last criterion is often preferred because it includes in the
analysis process the bene®ts to the users that are generated by the improvement in pavement
condition, and not just the M&R costs incurred by the agency. This provides a fairer assessment of
the value of each alternative (Novak and Kuo, 1987). There are also several ways in which the
analysis handles the timing decision it involves: the rehabilitation may be timed in accordance with
some redetermined trigger point (as is most often the case) or left to be determined simultaneously
with the treatment selection, a much more complicated task. A detailed discussion on life-cycle
cost analysis of pavements, as well as an extensive list of state agencies that use them, can be found
in Peterson (1985). This type of approach is also used in the PAVER system, but on an optional
basis (Shahin, 1983). It should be noted that this method relies heavily on the prediction models of
the system within which it is implemented and therefore its accuracy is strongly conditioned by
their own.
5.2.3. Dynamic programming. Dynamic programming, a well known optimization technique, can
also be used at this planning level. In particular, it is the natural tool to be used when resorting to
Airport pavement management systems
207
a Markovian model to predict future pavement performance (see section 4.2) but it can also be
used and be very e€ective in the deterministic context. The planning problem one faces in airport
pavement management, both at the project and network levels, is a problem where decisions are
made in stages with the objective to minimize a certain cost (or similarly to maximize a certain
bene®t). The outcome of each decision may not be fully predictable, the pavement degradation
process being, as was already mentioned, somewhat uncertain. However, since decisions are taken
in stages, the outcome of the present decision can be observed before the next stage (e.g. period or
year) decision is made and thus in¯uences it. A key aspect of such problems is that decisions cannot be viewed in isolation, since one must balance the desire of low present costs (in this stage)
with the possibility of high future costs being inevitable because of previous decisions. For
instance, delaying a rehabilitation (until some PCI value is reached, for example) might result in
lower costs for the initial periods of the planning horizon but it might also render necessary a more
extensive and costly rehabilitation action than an alternative strategy, in which rehabilitation is
performed earlier and followed by maintenance only for a lesser total cost. This idea is captured in
the dynamic programming technique, whereby at each stage one selects a decision that minimizes
the sum of the current stage cost and the best cost that can be expected from future stages. For
more details on dynamic programming, the reader is referred to Denardo (1982) and Bertsekas
(1987).
The decision that is made in this context is more far-reaching than with the previous methods.
Indeed, the dynamic programming approach enables the manager to explicitly take into account
the sequential nature of the decision at hand and therefore to simultaneously address both the
choice and timing decisions, something the previous methods do not really permit. This approach
determines the most cost-e€ective sequence of treatments to apply throughout the pavement life.
Not only does it determine an optimal chain of M&R actions, but it can also create, quite easily, a
set of near-optimal or non-optimal chains, which can be of use later, at the network level, when
dealing with budget constraints (Cook and Lytton, 1987).
This is, by far, the more powerful of the methods that may be used in project planning and can
handle many more potential M&R alternatives than the life-cycle analysis approach in a fraction
of the time. It has been slowly gaining acceptance and adaptations of this approach aimed at the
PAVER system in particular have been developed (Artman et al., 1983; Feighan et al., 1987; Butt,
1991) and tested with success on a real airport case (Butt, 1991).
5.2.4. Expert system. Although these types of methods have existed for some time in other contexts, they have just started to appear in PMSs. One such knowledge-based system called AIRPACS, for Air®eld Pavement Consultant System (Seiler, 1991; Seiler et al., 1991), has recently
been developed to solve dicult jointed plain concrete pavement (JPCP) design problems by using
the experience of planners, constructors, air®eld managers and designers. The system determines
the di€erent M&R alternatives to be considered based on traditional engineering practices that
generally rely on judgment (in fact, using the same guidelines as the PAVER system; see Hall et al.,
1992). These alternatives are then compared using life-cycle cost analysis methods using the usual
annualized cost as the economic criterion and the one minimizing this criterion is selected. During
validation testing, the recommendations of AIRPACS compared favourably with results obtained
by using current empirical and mechanistic design procedures.
These techniques are still in their infancy in the pavement management context. They are presently rather limited in scope, since they can only address design problems for some types of
pavement structures. In addition, it is not at all certain that they may ever be able to adequately
address complex planning decisions unless they venture beyond the knowledge and expertise
embodied in traditional methods and begin to incorporate the powerful optimization approaches
as well. Nevertheless, it may very well be that at some future time, systems of this sort will be an
integral part of many PMSs.
5.3. Network level planning
Network level planning is a problem in which decisions must be taken on numerous projects
simultaneously. While at the project level tradeo€s between projects and budget limitations were
not at issue, these two factors take on paramount importance in the analysis of the network as a
whole. It is, in fact, these two features which create the greater complexity inherent to the network
208
M. Gendreau and P. Soriano
planning problem when compared with project planning (Cook and Lytton, 1987). At this decision
level, one must distinguish between program planning and ®nancial planning. In program planning, decisions involve which projects are to be carried out, when they should be done and what
treatment should be used in order to maximize the overall network quality, while respecting
predetermined budgetary constraints. Financial planning, on the other hand, is concerned with
determining the level of funding required in order to maintain overall network quality (or health)
at some desired level. These two planning problems are complementary and are generally addressed separately. The approaches used to tackle the network planning problem are quite varied,
but they can be divided into two families: ranking methods, which are the traditional approach
to this problem, and optimization methods. We will now review these two management philosophies.
5.3.1. Ranking methods. Ranking methods, as we have just stated, are the traditional way in
which this problem is addressed. They are an extension of project planning to the network level
that belongs to the `from the bottom-up' philosophy of pavement management. Indeed, these
methods are sequential in nature with the scheduling of projects and the adherence to budget
constraints occurring after, and in isolation from, the M&R treatment selection and action year
decision (Cook and Lytton, 1987).
Hence this approach amounts to:
(a) Determine ®rst the M&R needs for each section in the network individually, generally using
traditional project-level techniques based on engineering judgment and/or life-cycle cost
analysis.
(b) Then, if the total M&R needs just identi®ed exceed the available budgets, a ranking of all
projects is established according to some decision criterion and the highest ranking projects
are sequentially selected to be carried out, up to the budget limit. Projects that cannot be
programmed in the current year are deferred to the next.
This procedure can be performed for a single year or for a multi-year planning horizon by
repeating it for each year. Several di€erent decision criteria may be used for ranking, depending on
the managing agency preferences, such as rank by distress, distress and trac, net present value,
bene®t±cost ratio, or other composite criteria relating to the particular characteristics and function
of each section.
This approach to network management planning is the most common in existing APMSs, as
well as in highway PMSs. They are found in both the PAVER (Shahin, 1983) and IAPMS (Rada
et al., 1992) systems. Ranking methods are conceptually simple, quickly implemented, and easy to
understand. However, they su€er from some very serious drawbacks. First of all, they do not yield
an optimal allocation of funds since the decisions they take are essentially project-level decisions
(made in isolation from one another) with the resulting network decisions really amounting to the
sum of a set of project decisions (Cook and Lytton, 1987). Furthermore, since the decisions are
taken sequentially without considering the overall network impact in the process, these methods
are unable to provide an estimation of the consequences for the network system of the decisions
taken. In fact they are reactive to the short-term needs, instead of being proactive in seeking
solutions to long-term needs (Butt, 1991). Finally, these methods are essentially a programming
tool and, thus, cannot adequately address ®nancial planning.
5.3.2. Optimization approaches. As was just seen, the ranking approach su€ers from several
limitations. This arises from the fact that it addresses network planning as a sequential selection
process of predetermined project-level decisions, which results in, as a main consequence, completely ignoring the potential inter- as well as intra-project tradeo€s. Unfortunately these are crucial aspects that need to be addressed in order to manage eciently at the network level. Hence,
the network planning problem should simultaneously consider all decisions pertaining to section,
M&R treatment, and timing selection, while adhering to the budget constraints. To do so, several
agencies have resorted to optimization approaches. These approaches require the formulation of
the decision problem as a mathematical model in which the objective one wishes to pursue (e.g.
minimizing total costs) and the constraints one has to satisfy (e.g. annual budgets, network performance requirements, such as minimum average condition or maximum number of sections in
Airport pavement management systems
209
unacceptable condition, operational restrictions, etc.) are stated as mathematical expressions of
the decision variables. This model can then be solved by one of several existing optimization
techniques.
The di€erent applications of optimization approaches found in the literature are set in the
highway context, but are perfectly usable in APMSs. These applications can be separated into two
distinct families depending on the way the individual sections composing the network are
accounted for. They can either all be explicitly considered, in this case we will have a detailed
model, or aggregated in classes or families, each corresponding to a group of sections having
similar characteristics. Both of these types of approach have their advantages and drawbacks.
In detailed models, such as those of Illinois (Mosheni et al., 1992), Ontario (Kher and Cook,
1985) and Danemark (Ullidtz, 1985) for example, the section are all individually identi®ed and,
therefore, the output from the optimization procedure directly provides an M&R program for the
whole planning horizon. Unfortunately, the size of the networks that have to be considered in
practice is such that the resulting mathematical models are often too large to be solved directly. To
overcome these problems one can resort to pre-processing techniques, in order to limit the number
of alternatives considered for each section as in Gendreau and Pehlivanidis (1985), Ullidtz (1985),
and Harper and Majidzadeh (1992), or use approximate solution techniques to solve the problem,
®nding, thus, good solutions but without being able to guarantee their optimality (see Gendreau
and Pehlivanidis, 1985; Kher and Cook, 1985; Ullidtz, 1985; Majidzadeh et al., 1992; Mosheni et
al., 1992). Often, both techniques must be implemented in order to suciently reduce the model
size. However, in the airport management context this problem should not be as severe since the
size of the networks is much smaller. It should be noted that these types of models can address
both programming and ®nancial planning considerations by simply modifying the objective function of the mathematical model.
In the case of aggregate models, the ®rst advantage is, of course, the considerable reduction in
size which allows one to optimally solve practical planning problems of huge dimensions on microcomputers, as is the case with the Finnish PMS (Thompson et al., 1987, 1989). These types of
models also provide managing agencies with a good overall image of the health of the network and
of its evolution with time. Markov models, or more exactly Markov decision processes, fall into
this category. These models, which are the natural extension of Markovian prediction models at
this planning level, have been applied with very impressive results in the Arizona PMS (Golabi et
al., 1982) generating savings of 14 million dollars (almost 1/3 of Arizona's prevention budget) in
the ®rst year of implementation alone, as well as in that of Finland (Thompson et al., 1987). The
major drawback these models present is the loss of identity of individual sections. As a result, the
translation of the optimal M&R network strategy into a workable program of speci®c M&R
projects is not trivial. This approach is therefore primarily a ®nancial planning approach and not
directly applicable program planning.
One way to deal with this limitation is to use heuristic rules and spreadsheet technology to
manually determine a feasible programming of M&R projects which adheres to the network
guidelines (Thompson et al., 1989). This implies a loss of optimality, but since the optimal network
guidelines are followed, the resulting solution should still be good. Another way of dealing with
this has been proposed in Gendreau (1987) and involves the separation of the overall network
planning problem into two complementary problems: a strategic planning problem (aggregated)
that essentially addresses the ®nancial planning aspect, and a tactical planning problem (detailed),
which uses the strategic solution as an input in order to determine a precise program of M&R
projects for each district or region independently.
Generally, the optimization techniques used for solving these models are either optimizationbased heuristics, linear programming, dynamic programming, or integer programming codes. As
can be seen, these approaches draw heavily on techniques borrowed from the operations research
®eld which, until recently, had got little attention from the engineering community. Although
more complex and abstract in nature, these approaches are much more powerful than any of the
traditional ranking methods as evidenced by the study conducted with the ILLINET system,
which illustrated the superiority of rather simple optimization approaches over ranking methods
(Mosheni et al., 1992). For instance, when comparing the rehabilitation programs obtained by a
classical ranking approach based on the worst-®rst selection rule (with respect to pavement condition) where the project available for each section was determined through life-cycle cost analysis
210
M. Gendreau and P. Soriano
and by solving a mathematical model that includes all project-level options using a linear programming method, one ®nds the following. Both methods produce programs having similar total
costs of 73.8 and 71.2 million dollars for the ranking and optimization approaches, respectively.
However, the user bene®ts generated by these programs, as measured by the number of vehicle
miles traveled on pavements in adequate condition, was much lower with the ranking method
program at 3.82 billion, than with the optimized program, where it attained 5.63 billion, an
increase of 47%. All of this plus the tremendous success of the Arizona PMS and the in¯uence it
has had on recently developed PMSs (Kulkarni, 1985; Thompson et al., 1987; Harper and Majidzadeh, 1992) are a clear indication of the growing appeal of optimization techniques and of the
huge bene®ts they can provide.
6. OTHER FEATURES
As with any other decision-support software, APMSs are systems that must provide facilities to
integrate the various aspects of the decision-making process into a consistent framework: input/
output procedures, data management capabilities, etc. In many ways, these facilities are critical to
the smooth operation of an APMS and to its e€ectiveness as a management tool. We will now
brie¯y review the most commonly encountered of these features.
6.1. User interface
The user interface is the front end of the APMS. It is through this feature that the user communicates and interacts with the system to access the database, run the algorithms, prepare
reports, etc. The user interface de®nes the work environment proposed by the APMS to its
potential users. The sophistication of this element can vary signi®cantly from one system to another,
ranging from quite simple functions to very elaborate programs. The more advanced of these user
interfaces, as seen in ERES Consultants (1993) and PCS/Law Engineering (1993), constitute a very
user-friendly environment that is generally interactive, fully menu-driven and makes extensive use
of high-resolution color graphics. They often provide, in addition, an extensive on-line help function that is context sensitive. This particular facility reduces the strain of the training process for
new users and minimizes the need for time-consuming consultation of hard-copy manuals.
Some systems, like the IAPMS (Schwartz et al., 1991; Rada et al., 1992), have integrated concepts drawn from the ®eld of geographic information systems (GISs) with the graphical capabilities of their user interface and their database (i.e. incorporating spatial information to the data).
This allows the user to perform spatial as well as conventional queries and analyses of the database contents. It also enables the results of these analyses or queries to be displayed in a geographical context, thus providing a di€erent and more global perspective to interpret them. This is an
important aspect, not only to engineering sta€, but also to management personnel. A further
advantage of these GIS-graphical capabilities is their ability to fully integrate the wide range of
formats on which the information required for pavement management is generally reported (i.e.
maps, drawings, charts, tables, etc.). More detailed discussions of GIS concepts as they apply to
PMS can be found in CoÃte et al. (1991), Schwartz et al. (1991) and Murrell et al. (1992).
6.2. Data management facilities
The database is the cornerstone of the APMS. It serves as the repository of all information
pertaining to the decision-making process. Given the extent of historical data and the high level of
detailed information required for this task, the size of these databases is generally quite large. In
addition, these data are strongly hierarchical in nature (e.g. construction cost and M&R policies
are global information that pertain to the airport as a whole, while detailed distress observations
or NDT results are data associated with a speci®c pavement section). Finally, new data are regularly collected and added to the database.
All this creates special needs that must be addressed when developing the database structure and
its management system. In particular,the following characteristics should be satis®ed (Hudson and
Uddin, 1987):
(a) ease of data entry, editing and validation;
(b) ease and eciency of data reduction;
Airport pavement management systems
211
(c) eciency, ¯exibility, and quality of database query and reporting facilities;
(d) adaptability to future enhancements and extensions.
Most data management facilities found in APMSs are organized in a hierarchical indexed
structure (Shahin, 1983; Rada et al., 1992) to re¯ect the similar nature of the data and to speed up
manipulations. They also generally satisfy the requirements listed above.
6.3. Report generation capabilities
Another important feature of APMSs is the way in which the results of the di€erent analyses it
can perform are assembled and presented to the user. In order to maximize the usefulness of its
analysis capabilities, the APMS should be able to generate di€erent types of reports that address
the di€erent needs and expectations of the various management levels involved. There should
minimally be three levels of detail in these reports, as in the case in Mosheni et al. (1990):
(a) network summary, providing a synthesis of the results for the network as a whole;
(b) project summary, containing the rehabilitation programs and costs for each section in the
network;
(c) project detailed, containing all the detailed data or results for each section of the network.
Most systems provide these minimal capabilities and generally more.
Another very useful format in which results can be presented is through the use of graphics:
histograms, pie charts, time plots, and color-coded maps, among others. These graphical report
capabilities constitute the ideal complement to the traditional report format by providing a clear
and concise representation of the information. Furthermore, this representation lends itself extremely well to scenario analyses of the `what if'-type questions by enabling the results of di€erent
alternatives to be displayed on the same graphics or in the exact same format, which allow quick
and easy comparisons. The IAPMS (Rada et al., 1992; PCS/Law Engineering, 1993) and ERES
DSS (ERES Consultants, 1993) provide such report capabilities.
6.4. Decision-support environment
As was mentioned at the beginning of this section, a modern APMS is ultimately a decisionsupport system and as such its role is to serve as a tool and aid to decision-making rather than as a
decision-maker. Given the multi-objective nature of the management problem at hand, in which
one can try to maximize bene®ts or network health, minimize costs or any of several other objectives, it is essential that the system provide the planners with a minimum range of interaction
facilities in order to enable them to perform meaningful analyses and comparisons. It is through
these facilities that the numerous `what if' questions that arise can e€ectively be addressed.
The decision-support environment, as these facilities are often referred to, should provide the
user with the ability to:
(a) modify the output and quickly evaluate the consequences of these modi®cations (®nancial
as well as physical);
(b) test the sensitivity of the proposed solutions to alternative options or constraints;
(c) de®ne various objectives in order to analyze di€erent aspects of the problem and compare
them.
Although most existing systems do provide some facilities for user interaction, these are, for the
most part, somewhat limited in scope and nature. To our knowledge, they essentially address the
®rst type of user interaction listed above and not the others.
6.5. Modularity, adaptability, ¯exibility
A ®nal feature of considerable value in any APMS is its ability to grow and adapt to the evolving needs of the agency that uses it. First of all, it should be noted that the full and complete
implementation of such a system is a process that is usually stretched over a signi®cant period of
time (i.e. a few years) and which is often accomplished in several disjoint phases. Thus, the modularity of the system and the ability of its component modules to be implemented and used
somewhat independently one from another, is an important asset that greatly facilitates the overall
implementation process and eases the adjustment of the users to the new tools.
212
M. Gendreau and P. Soriano
A second desirable quality in the design of such a system would be the ability to adapt and
update its component modules with the changing needs of the using agency and the growing
amounts of detailed data available. For instance, the prediction module should ideally be able to
integrate all new condition data as they become available and, therefore, provide the most accurate performance predictions possible.
Finally, the overall organization of the system should be such that it has the necessary ¯exibility
and adaptability to enable the modi®cation of some module (e.g. upgrading the network planning
function from ranking methods to optimization techniques) without altering the rest of the system.
In this way, the APMS will be able to continuously take advantage of the most up-to-date powerful methods available to carry out its main functions.
7. CONCLUSION
This paper has presented a description of the essential components making up present airport
pavement management systems. These are network inventory; pavement condition evaluation,
performance prediction models, and planning methods. For each of these elements, the various
techniques used in the di€erent PMSs developed for the management of air®eld networks have
been reviewed. We have also included in this review several methods implemented in highway
PMSs because most of the concepts used in this ®eld can generally be transferred to the airport
management context without diculty.
The ®rst observation that can be drawn from this work is that the variety of methods implemented in highway PMSs is far greater than those encountered for airport PMSs, where a particular management approach, namely the PAVER system, has almost become the standard way of
proceeding. This situation does not appear to be related to articular technical factors speci®c to
the air®eld pavement context, but more to historical reasons.
In spite of this, the new techniques and approaches for managing pavements that have been
developed in the highway context could very well be applied to air®elds. Among these, probabilistic performance prediction models, in particular Markovian models, dynamic programming
approaches to project level planning, and optimization models for network level planning seem to
o€er the greatest potential for enhancing the capabilities of APMSs. In fact, adaptations of several
of these techniques within the PAVER framework have already been proposed and successfully
implemented.
As of now, it is yet unclear whether these extensions to the PAVER approach will be sucient
to meet the expectations of management with respect to APMSs in the coming years. It is quite
possible that other approaches, in particular those based on the `from the top-down' philosophy,
will prove more successful in addressing these needs and ultimately become the standard way of
managing air®eld pavements.
AcknowledgementÐThis research was funded by AeÂroports de MontreÂal.
REFERENCES
Alexander, D. R. and Hall, J. W. (1991) ACN-PCN concepts for airport pavement management. In ASCE Conf. on Aircraft/Pavement Interaction: An Integrated System, Kansas City, September 393±405.
Antunes, M. L. and Pinelo, A. (1990) Airport pavement evaluation and ACN-PCN classi®cation. In Proc. 3rd International
Conf. on Bearing Capacity of Roads and Air®elds, 2:1019±1029, Trondheim, Norway, July.
Artman, D. H. Jr., Liebman, J. S. and Darter, M. I. (1983) Optimization of long-range major rehabilitation of air®eld
pavements. Transportation Research Record 938, 1±11.
Baladi, G. Y., Novak, E. C. Jr. and Kuo, W. H. (1992) Pavement condition index remaining service life. In Pavement
Management Implementation, eds F. B., Holt and W. L., Gramling, STP 1121, American Society for Testing and
Material, Philadelphia, PA, pp. 63±90.
Bertsekas, D. P. (1987) Dynamic Programming: Deterministic and Stochastic Models. Prentice-Hall, Englewood Cli€s, NJ.
Butt, A. A. (1991) Application of Markov process to pavement management systems at the network level. Ph. D. Dissertation, Department of Civil Engineering, University of Illinois at Urbana-Champaign.
Cation, K. A. and Schwandt, G. (1991) Automating air®eld condition data collection. In Proc. ASCE Conf. on Aircraft/
Pavement Interaction: An Integrated System, Kansas City, September, 377±392.
Cook, W. D. and Kazakov, A. (1987) Pavement peformance prediction and risk modelling in rehabilitation budget planning: a Markovian approach. In Proc. Second North American Conf. on Managing Pavements, 2:2. 632. 75, Toronto,
Canada, November.
Airport pavement management systems
213
Cook, W. D. and Lytton, R. L. (1987) Recent developments and potential future directions in ranking and optimization
procedures for pavement management. In Proc. Second North American Conf. on Managing Pavements, 2:2. 135±2.
155, Toronto, Canada, November.
CoÃteÂ, J., Roy, S. and Rousseau, J. -M. (1991) SysteÁme graphique d'aide aÁ la gestion de l'entretien du reÂseau routier d'une
municipaliteÂ. Publication CRT-765, Centre de recherche sur les transports, Universite de MontreÂal.
Darter, M. I., Smith, R. E. and Shahin, M. Y. (1985) Use of life cycle cost analysis as the basis for determining the coste€ectiveness of maintenance and rehabilitation treatments for developing a network level assignment procedure. In
Proc. North American Pavement Management Conf., 2:7.5±7.18, Toronto, Canada, March.
Denardo, E. J. (1982) Dynamic Programming. Prentice-Hall, Englewood Cli€s, NJ.
Eckrose, R. A. and Reynolds, W. G. (1992) Implementation of a pavement management system for Indiana airports A case
history. In Pavement Management Implementation, eds F. B. Holt and W. L. Gramling, STP 1121, American Society for
Testing and Material, Philadelphia, PA, pp. 228±239.
ERES Consultants Inc. (1993) Statement of quali®cations of ERES DSS for pavements and AIRPACS.
Federal Aviation Administration (1980) Procedure for condition survey of civil airports. Report FAA-RD-80-55,
Washington, DC.
Feighan, K. J., Shahin, M. Y. and Sinha, K. C. (1987) A dynamic programming approach to optimization for pavement
management systems. In Proc. Second North American Conf. on Managing Pavements, 2:2.195±2.206, Toronto,
Canada, November.
Gendreau, M. (1987) A decomposition approach for rehabilitation and maintenance programming. In Proc. Second North
American Conf. on Managing Pavements, 2:2.207±2.218, Toronto, Canada, November.
Gendreau, M. and Pehlivanidis, M. (1985) A heuristic for the multi-period programming of structural rehabilitation projects. In Proc. North American Pavement Management Conf., 2:6.55±6.66, Toronto, Canada, March.
Golabi, K., Kulkarni, R. B. and Way, G. B. (1982) A statewide pavement management system. Interfaces 12, 5±21.
Hall, J. W., Grau, R. W., Grogan, W. P. and Hachiya, Y. (1992) Performance indicators from army air®eld pavement
management program. In Pavement Management Implementation, eds F. B. Holt and W. L. Gramling, STP 1121,
American Society for Testing and Material, Philadelphia, PA, 297±317.
Harper, W. V. and Majidzadeh, K. (1992) Innovations in PMS State of Ohio & Kingdom of Saudi Arabia. In Pavement
Management Implementation, eds F. B. Holt and W. L. Gramling, STP 1121, American Society for Testing and Material, Philadelphia, PA, 359±375.
Herrin, S. M., Darter, M. I., Barenberg, E. J. and Shahin, M. Y. (1986) Air®eld pavement evaluation and management at
Dulles international airport. Transportation Research Record 1060, 53±61.
Hudson, S. W., Hudson, W. R. and Carmichael, R. F. (1992) Minimum requirements for standard pavement management
systems. In Pavement Management Implementation, eds F. B. Holt and W. L. Gramling, STP 1121, American Society for
Testing and Material, Philadelphia, PA, 19±31.
Hudson, W. R. and Uddin, W. (1987) Future pavement evaluation technologies: prospects and opportunities. In Proc.
Second North American Conf. on Managing Pavements, 2:3;233±3.258, Toronto, Canada, November.
Kher, R. K. and Cook, W. D. (1985) PARS the MTC model for program and ®nancial planning in pavement rehabilitation.
In Proc. North American Pavement Management Conf., 2:6.23±6.40, Toronto, Canada, March.
Kulkarni, R. B. (1985) Development of performance prediction models using expert opinions. In Proc. North American
Pavement Management Conf., 1:4.135±4.147, Toronto, Canada, March.
Lary, J. and Soule, R. (1991) DFW pavement management system implementation. In Proc. ASCE Conf. on Aircraft/
Pavement Interaction: An Integrated System, Kansas City, September, pp. 406±420.
Lee, H. (1992) Standardization of distress measurements for the network-level pavement management systems. In Pavement
Management Implementation, eds F. B. Holt and W. L. Gramling, STP 1121, American Society for Testing and Maerial,
Philadelphia, PA, 424±436.
Lee, K. W. and Bowen, G. E. (1992) Standardization in pavement management implementation for municipally maintained
roads in Rhode Island. In Pavement Management Implementation, eds F. B. Holt and W. L. Gramling, STP 1121,
American Society for Testing and Material, Philadelphia, PA, 211±227.
Lee, H. and Hudson, W. R. (1985) Reorganizing the PMS concept. In Proc. North American Pavement Management
Conf., 1:2.59±2.80, Toronto, Canada, March.
Livneh, M. (1988) Comments on modi®ed models for maintenance and rehabilitation planning. Transportation Research
Record 1200, 42±52.
Lytton, R. L. (1987) Concepts of pavement performance prediction and modeling. In Proc. Second North American Conf.
on Managing Pavements, 2:2.5±2.12, Toronto, Canada, November.
Majidzadeh, K., Saraf, C. L. and Kennedy, J. C. Jr. (1992) Ingredients of a third generation pavement management system
for the Ohio department of transportation. In Pavement Management Implementation, eds F. B. Holt and W. L.
Gramling, STP 1121, American Society for Testing and Material, Philadelphia, PA, 318±333.
Monismith, C. L. (1977) An overview of air®eld pavement design. In Proc. ASCE Air Transportation Division Special
Conf. of the Institute of Air Transportation, Washington, DC, April, 256±324.
Monismith, C. L. (1978) Considerations in airport pavement management. Transportation Research Board, Special Report
175, 10±34.
Mosheni, A., Darter, M. I. and Hall, J. P. (1990) Illinois pavement network rehabilitation management program. Transportation Research Record 1272, 85±95.
Mosheni, A., Darter, M. I. and Hall, J. P. (1992) E€ect of selecting di€erent rehabilitation alternatives and timing on network performance. In Pavement Management Implementation, eds F. B. Holt and W. L. Gramling, STP 1121, American
Society for Testing and Material, Philadelphia, PA, 117±131.
Murrell, S. D., Rada, G. R. and Schwartz, C. W. (1992) Airport pavement management: the port of New York and New
Jersey experience. In Pavement Management Implementation, eds F. B. Holt and W. L. Gramling, STP 1121, American
Society for Testing and Material, Philadelphia, PA, 240±255.
Novak, E. C. and Kuo, W. H. (1987) A comparison of life-cycle cost and cost e€ective methods of economic analysis. In
Proc. Second North American Conf. on Managing Pavements, 3:3.71±3.80, Toronto, Canada, November.
Paterson, W. D. O. and Robinson, R. (1992) Criteria for evaluating pavement management systems. In Pavement Management Implementation, eds F. B. Holt and W. L. Gramling, STP 1121, American Society for Testing and Material,
Philadelphia, PA, 148±163.
214
M. Gendreau and P. Soriano
PCS/Law Engineering (1993) Development and implemntation of an airport pavement management system (APMS). Ref.
No. P930091, July.
Peterson, D. E. (1985) Life-cycle cost analysis of pavements. NCHRP Synthesis 122, Transportation Research Board,
Washington, DC.
Rada, G. R., Schwartz, C. W., Witczak, M. W. and Rabinow, S. D. (1992) Integrated pavement management system for
Kennedy International Airport. Journal of Transportation Engineering 118, 666±685.
Rooke, W. G. (1985) Pavement management system. Transportation Research Forum 1(4), 12±16.
Ross, S. M. (1970) Applied Probability Models with Optimization Applications. Holden-Day, San Francisco, CA.
Schwartz, C. W., Rada, G. R., Witczak, M. W. and Rabinow, S. D. (1991) GIS applications in air®eld pavement management. Transportation Research Record 1311, 267±276.
Seiler, W. J. (1991) A knowledge-base for rehabilitation of air®eld concrete pavements. Ph. D. Dissertation, University of
Illinois.
Seiler, W. J., Darter, M. I. and Garrett, J. H. Jr. (1991) An air®eld pavement consultant system (AIRPACS) for rehabilitation of concrete pavements. In Proc. ASCE Conference on Aircraft/Pavement Interaction: An Integrated System.
Kansas City, September, 393±405.
Shahin, M. Y. (1980) Components of a pavement management system. Transportation Research Record 781, 31±39.
Shahin, M. Y. (1982) Air®eld pavement distress measurements and use in pavement management. Transportation Research
Record 893, 59±63.
Shahin, M. Y. (1983) Airport pavement management A total system. International Air Transportation Conf., Montreal,
Canada, June.
Shahin, M. Y. (1992) 20 years experience in the PAVER pavement management system: Development and implementation.
In Pavement Management Implementation, eds F. B. Holt and W. L. Gramling, STP 1121. American Society for Testing
and Material, Philadelphia, PA, 256±271.
Shahin, M. Y. and Becker, J. M. (1984) Development of performance prediction models for air®eld pavements. Transportation Research Record 985, 253.
Shahin, M. Y., Darter, M. I. and Kohn, S. D. (1979) Evaluation of air®eld pavement condition and determination of
rehabilitation needs. Transportation Research Record 700, 1±10.
Shahin, M. Y. and James, T. D. (1984) Development of a pavement maintenance management system, volume X: Summary
of development from 1974 to 1983. Technical Report, Air Force Engineering and Services Center, Engineering and
Services Laboratory, Tyndall AFB, FL.
Shahin, M. Y. and Kohn, S. D. (1982a) Overview of PAVER. Transportation Research Record 846, 55±60.
Shahin, M. Y. and Kohn, S. D. (1982b) Air®eld pavement performance prediction and determination of rehabilitation
needs. In Proc. 5th International Conf. on the Structural Design of Asphalt Pavements, 1:637±652, Delft, The Netherlands, August.
Tessier, G. -R. (1990) Guide de construction et d'entretien des chausseÂes. Association queÂbeÂcoise du transport et des routes,
MontreÂal, Canada.
Thompson, P. D., Neumann, L. A., Miettinen, M. and Talvitie, A. (1987) A micro-computer Markov dynamic programming system for pavement management in Finland. In Proc. Second North American Conf. on Managing Pavements,
2:2.241±2.252, Toronto, Canada, November.
Thompson, P. D., Olsonen, R., Talvitie, A. and Tapio, R. (1989) A micro-computer Markov model for optimal pavement
rehabilitation policy. In Proc. 5th World Conference on Transportation Research, Yokohama, Japan.
Ullidtz, P. (1985) A Danish pavement management system. In Proc. North American Pavement Management Conf.,
2:6.84±6.95, Toronto, Canada, March.
Ullidtz, P. and Larsen, B. K. (1983) Mathematical model for predicting pavement performance. Transportation Research
Record 949, 45±55.
Way, G. B. (1985) Network Optimization System for Arizona. In Proceedings of the North American Pavement Management Conf., 1:6.16±6.22, Toronto, Canada, March.
Witczak, M. W. (1978) Framework for evaluation and performance of airport pavements. Special Report 175, Transportation Research Board, 69±75.
Zaniewski, J. (1991) Uni®ed methology for airport pavement analysis and design, Vol. I State-of-the-Art. Technical Report,
Federal Aviation Adninistration, Research and Development Service, Washington, DC.
Descargar