| Return to Main
Page Design of a Long Term Pavement Monitoring System for the Canadian Strategic Highway Research Program
PHASE II: AUTOMATED DELIVERY OF
CLTPP RESULTS TO PROVINCIAL AND FEDERAL GOVERNMENT
AGENCIES Task II.1. Implement
Automated Output to Support Deterministic Pavement
Deterioration Functions. Once the Bayesian
statistical module has been implemented and integrated
into the statistical package, the first step is to
deliver the pavement performance function f(x,b*) and the
model parameters b* to pavement decision makers. This
information must be delivered in unambiguous, automated
form so that pavement management systems based on
deterministic methods can use it. Section 3.1 outlined
the basic structure of such systems and noted how they
are driven by pavement performance prediction
relationships. It will also be important to provide a
selfdocumenting format for delivering the pavement
deterioration functions and parameters resident within
the statistical package. In particular, the
userfriendly features of the statistical package will
have to be used to report the pavement deterioration
function and parameters. If insufficient userfriendly
features exist, it may be necessary to design and build a
simple reporting module and append it to the statistical
package. This task will require three (3)
calendar months and approximately 3 manmonths to
implement. Task II.2. Implement
Automated Output to Support SemiMarkovian Probabilistic
Pavement Deterioration Functions. The next task is to
implement automated output capability to report the
posterior probability distribution over pavement
performance determined using Bayesian statistical
methods. The process will be parallel to the process of
developing the reporting capability for the deterministic
deterioration relationships. This task will require two (2) calendar
months and approximately 2 manmonths to implement. Task II.3. Implement Remote Access
Procedure for Provincial and/or State Government Agencies
to Access Pavement Deterioration Information. A
critically important feature of the CLTPP system will
be the hardware, software, protocol, and standards by
which remote users can access the system. This task
involves designing and implementing ports by which the
CLTPP results can be accessed remotely and
automatically by pavement management personnel in the
provinces and states. Included will be analysis of
alternative communication ports, alternative
communication software and protocol, development and
documentation of communication standards, design of
specific information to be accessed, and design of
specific information to be received from remote users. A critically important aspect of remote
access will be to ensure security and integrity of the
CLTPP data and modules. The remote access capability
should contain unequivocal, stateoftheart
procedures and management guidelines for protecting the
security and integrity of the data and models to which
users will be provided access. Another important aspect of security
will be to design backup capability for the CLTPP data
base. Furthermore, CLTPP must implement a "master
data base" updating procedure so that a single,
inviolate copy of the master data base is accessible only
by a small number of cognizant individuals via password
access. This master data base should allow
"readonly" access to users lacking the
password. This task will require 36 calendar
months to complete and approximately 612 manmonths to
implement. PHASE III: OPERATION IN
CONJUNCTION WITH INTEGRATED, PROBABILISTIC PAVEMENT
DESIGN AND MANAGEMENT CAPABILITY This phase of the project is
sufficiently far in the future that we will not present
detailed time or level of effort estimates herein. Task III.1. Acquire
SemiMarkovian Module for SectionSpecific Pavement
Management. As articulated in Section 3.2 and
Appendix A, the most sophisticated and most accurate
method for pavement design and management is the
probabilistic semiMarkovian approach. Following
development of the Bayesian statistical system and the
automated access and retrieval system in Phases I and II,
it will then be time for CLTPP to implement the
decision making capability that can use that information
to optimally manage the highway system. The first step,
which is the subject of this task, is for CLTPP to
acquire access to the fundamental building blocka
single section semiMarkovian pavement design and
management module. Clayton Sparks and Associates and
Decision Focus Incorporated have developed such a module
and offer it on a license basis at modest cost. (The
trade name of our product is PIMS, which stands for
Pavement and Infrastructure Management System.) As far as
we know, the Clayton Sparks/DFI PIMS module is the only
such module in existence. Furthermore, our module has
been extensively tested and proven on pavement sections
in Saskatchewan and Manitoba. We are confident that our
offering will be extremely cost effective relative to the
alternative of designing one from scratch. Whatever semiMarkovian module is
selected, complete methodological and user documentation
should be included in this task. Task III.2. Incorporate User Cost
Calculation Capability. The semiMarkovian approach
is rather uniquely capable of considering user costs as
well as agency costs in determining the optimal design
and maintenance strategy for a pavement section (and
through integration for the highway system as a whole).
The purpose of this task will be to assign realistic user
costs to each section of pavement for each performance
state in which that pavement can reside. Clayton Sparks
has made such assessments both in support of the Manitoba
application of our semiMarkovian capability as well as
in support of other transportation studies throughout
North America. As articulated in Appendix A, inclusion of
user cost is a critical link in understanding the proper
application of highway funds and even more importantly in
advocating and justifying additional highway funds from
legislative funding bodies. As we have discussed, the
semiMarkovian approach is capable of measuring the
expected net benefits to users for every possible
pavement design and maintenance strategy and to select
the optimum strategy. The optimum strategy is the one
that maximizes the expected net benefits minus agency
costs. (As a trivial special case, we can minimize
expected agency costs necessary to maintain a given level
of service.) This task involves the estimation of
user costs and agency costs for every section in the
system and implementation within the semiMarkovian
module. Cofunding by individual provincial jurisdictions
is anticipated before this task can be effectuated. Task III.3. Interconnect Multiple
Section Modules into District and System Pavement
Management Systems. Once the sections are represented
using the semiMarkovian module, the next step is to
bind all the modules together through budget constraints
to form a model of provincial districts and of the
provincial system as a whole. This model of the system as
a whole will allow users to solve what we believe to be
the most difficult and perplexing problems in the
pavement system: If budget dollars are scarce, where in
the system should they be applied? Interstates?
Arterials? Should certain sections be
"goldplated" while other sections are
ignored? What if CLTPP data begin to imply different
pavement deterioration mechanisms and rates than are
currently believed or are embedded in current practice?
How should decision makers adapt their practices? What
should they do differently? This latter question is perhaps the
most difficult. How should provinces and states change
what they do as definitive new CLTPP data comes in? The
district and system level semiMarkovian models give
precise answers. Task III.4. Design Advanced
Graphical and Report Writing Capability. Once the
semiMarkovian models have been interconnected to form a
district or system wide model, it will be important to
implement advanced graphical and report writing
capability so that decision makers can clearly understand
and communicate the basis for the pavement design and
management strategies emanating from Task III.3. This
will be critically important when we move to Task III.5,
which extracts the insights from the pavement design and
management systems and presents is using simple
principles or rules of thumb. The specific nature of the report
writing capability will depend on the number of
semiMarkovian modules integrated into a system model
and the particular nature of the budget constraint. Task III.5. Develop
Pavement Management "Principles" Consistent
with the Best Available Data and SemiMarkovian Pavement
Management System. As experience is gained with the
systemwide semiMarkovian model in Task III.3,
consistent principles of prudent pavement management will
begin to emerge. By this we mean that similar strategies
that apply to a broad range of pavements in a broad range
of environments will begin to emerge. The objective of
this task will be to identify those commonalities,
document them in simple terms, document their rationale
in simple terms, and deliver them as simple operating
principles or rules of thumb to pavement managers in the
field. Task III.6. Develop
Training Program for Initial and Ongoing Training of
Users. Once the system is entirely implemented, it is
important to develop a training program for the system as
a whole. To do so, we will expand the training program
articulated in Task I.8 for the Bayesian statistical
module to incorporate the pavement management element in
concert with the Bayesian statistical element. The
training program will be initially delivered by DFI and
Clayton Sparks. As it is routinized, it will be given by
inhouse staff on an ongoing basis. Task III.7. Implement
and Manage Effective User's Group. One of the most
effective methods of ensuring ongoing support and service
is to establish a User's Group. DFI and Clayton Sparks
will organize and manage a User's Group for the Bayesian
statistical module, the semiMarkov module, and the
integrated system as a whole. Semiannual meetings at
disparate locations throughout Canada would be held.
Invited speakers from among the User's Group would
present topical results and problems at such meetings.
Improvements and enhancements would be defined,
discussed, refined, and commissioned by the User's Group
at their discretion. A User's Group newsletter and
updated user documentation would emanate from User's
Group meetings. It is important to relate what parts of
the proposed pavement monitoring, performance prediction,
and pavement management system already exist and what
parts must be built from scratch. The remainder of this
section outlines the two key portions of the system that
already exist (or can be easily adapted from existing
commercial packages) and the two key portions of the
system that must be constructed. The parts of the system that presently
exist include: 1. The data management system resident
in BMDP, SAS, SPSS, Oracle, Lotus, and perhaps equivalent
statistical packages. We recommend using whatever
preexisting statistical package that is most compatible
with a Bayesian statistical procedure (which does not yet
exist). DFI has used SAS successfully as a data
management system interfaced with several custom
applications programs to support some of its software
products. DFI has also written successful Oracle and
Lotus capabilities with significant probability and
statistical capabilities. 2. The semiMarkovian pavement
management software.3 The semi Markovian software has
been proven in applications in Manitoba and Saskatchewan. The parts of the system that need to be
built include: 1. The Bayesian statistical nucleus
that makes the calculations indicated in Section 2.3. The
requisite calculations are rather straightforward, but we
are unaware of any preexisting statistical packages in
which they exist. DFI and ClaytonSparks are uniquely
capable of building the Bayesian statistical capability
and interfacing it with a statistical package. Task I.4
and its predecessors characterized the necessary
activities. 2. The automated procedure that
delivers the pavement deterioration model and attendant
parameters from the statistical data management system to
any and all regional users of quantitative pavement
management system. Designing such a data management
system is not difficult but involves important
subtleties. DFI has built such interfaces before for
organizations such as Hertz and EPRI. 4.3 THE PAYOFFEXPLICIT
QUANTIFICATION OF THE VALUE OF INFORMATION Bayesian analysis in combination with probabilistic, semiMarkovian pavement management techniques allow the use of a fascinating and very valuable concept known as the value of information. In a nutshell, the value of information concept asks and answers the question: how much is the posterior distribution worth relative to the prior distribution? The answer to this question, which can
be obtained only with an inherently probabilistic
approach, involves the following specific steps: 1. What is the probability distribution
over benefits net of agency costs if pavement design and
management decisions are made optimally using only
the prior probability distribution? 2. What is the probability distribution
over benefits net of agency costs if pavement design and
management decisions re made optimally using the
posterior probability distribution determined after the
program data are in? 3. If the optimum design and management
strategy is the same under both the prior and posterior
distribution, then the benefits of the posterior
distribution relative to the prior will be zero. It will
have zero value because no pavement design and management
decisions will be affected. 4. If the optimum design and management
strategy is different under the prior and posterior
distributions, then the benefits of the posterior
distribution relative to the prior will be positive.
People will make different decisions as a result of the
pavement monitoring program, and those decisions will
save money and increase pavement effectiveness and user
benefits. The Bayesian approach in combination
with the semiMarkovian pavement management model
proposed here is uniquely able to calculate the value of
the information, expressed in dollars and cents, of the
information gathered on the long term pavement monitoring
program. The regression method will not be capable of
making such an assessment because it will be plagued by
small sample size caveats and qualifications. The value
of interim and final information assembled by our
proposed program is of centrally important value in
justifying the long term pavement monitoring program over
time to politicians and highway constituents. Our
approach will allow users of the long term pavement
monitoring data to quantify explicitly how much that data
is worth to them, and such quantification will begin with
the very first program data assembled. What is the alternative to a
systematic, justifiable measure of the value of the
information gained on the long term pavement performance
program? The answer appear to be "hand waving."
Quoting from a SHRP memorandum: "The general quantification of
potential LTPP benefits begins with the assumption that
10% extended pavement life and improved serviceability
level can be achieved from overlay and rehabilitation
alternatives as a result of LTPP research." There is no justification whatever;
this is a hip shot estimate devoid of any substantiation.
Why shouldn't they "assume" 20%? 50%? 75%? For
that matter, why not argue that LTPP will make highway
travel free? Assuredly, political bodies will require
more justification than this. Assuredly some or all
provincial governments asked to provide fifteen or more
years' of funding with no immediately discernible results
will ask for justification and accountability. None will
be forthcoming from hip shot responses of the foregoing
type. None will be forthcoming if a regression approach
coupled with a deterministic pavement management
framework is used. The Bayesian approach coupled with the
probabilistic pavement design and management approach
will allow you to quantify precisely and justify cogently
the economic benefit of improved pavement information,
and you will be able to do so over time beginning with
the initial data assembled on the program. |