1. INTRODUCTION
Forecasting deterioration of bridges is an important component in bridge management
system for optimising the maintenance phase. It helps to understand the current stage of
bridge condition and develop an economical decision to ensure sustained serviceability and
safety of the structure. VicRoad is in favour to develop the component extensively over
current use of level 2 bridge inspection pose high uncertainty. The condition rating scheme
is largely dependent on the selection bias that possible contain measurement errors which
tend to weaken the accuracy of the result. This review will explore the application of
forecasting bridges deterioration. The application includes the deterministic models,
artificial intelligence models and stochastic models focusing Markov approach method for
this investigation.
Deterministic models
A model using mathematical in which the outcomes represent through a known relationship
between the factors affecting facility deterioration and condition of the facility without the
availability of random variation. The model assumes that future bridge performance is
known with certainty. A commonly used deterministic model is the regression function
obtains by doing a regression analysis on historical bridge data (Veshosky et al 1994). This
model can pull limitation from predicting the average condition bypass the effect of
condition of individual’s facilities (Madnat 1995, Jang and Sinha 1989). The impact of
deterioration mechanisms of bridge deck and the deck joints are not in consideration of this
model (Sianipar and Adams, 1997). Deterministic model will trouble to update new data
because of high difficulty. Also it underestimates model error from the cause of the
optimization best fit and that is any misprediction indicates a model failure.
Artificial intelligence models
Artificial intelligence (AI) models perform computer information-processing to develop
forecast models inspired by the densely interconnected, parallel structure of human brain
process information (Aleksander and Morton,. 1990). AI interacts with range of systems,
artificial neural networks (ANN), genetic algorithm (GA), and case based reasoning (CBR). A
statistical analysis identify significant factors influencing the deterioration where treating
the deterioration as a pattern classification problem and eleven parameters of the decks are
identified to enhance the model(maintenance history, age, previous condition, district,
design load, deck length, deck area, environment, degree of skew, number of spans, average
daily traffic). An investigation has been made by Ying-Hua Huang (2000), propose fully
connected and layered feed-forward networks using back-propagation approach multilayer
perceptron (BP-MLP) classifier using the Artificial Neural Network prediction model for deck
deterioration. Using this method with the three ways cross validation from records of 942
concrete decks of Wisconsin database, prediction model reaches classification rate of
84.66% and 75.39% for training sets and the testing sets, respectively.

2. These results are highly significant that AI models adopt BP-MLP promise to be another
possible prediction tool.
Optimisation method follows the success of Artificial intelligence models was conducted
associate with high cost due to the computation process generating a large data output. An
introduction of hybrid optimisation method is to filter out feasible condition ratings as
optimisation scheme for Neural Network in predicting long term deterioration of the bridges
(Callow D, Lee J., Blumenstein M., Guan H., Loo Y. C. 2012). Case base reasoning is used
to reduce the number of possible cases for a bridge, but can still leave a large amount of
possibilities. A hybrid process consisting of both Case Base Reasoning – efficient in testing
process and Genetic Algorithm – high systematic processes will ensure that the possible
cases are optimised to their best potential. The development of the hybrid optimisation
method of Case Based Reasoning and Genetic Algorithm ensures consistently optimised
bridges with accurate future predictions of deterioration. The advancement of the hybrid
optimisation method using both GAs and CBR techniques as in the following:
 Optimisation is performed at a much early stage where unreliable condition ratings
are eliminated thereby reducing the source of uncertainties;
 When an optimised set of condition ratings is used as input values for the current AI-
based deterioration model, the prediction errors will be greatly reduced;
 The computational time required for long-term prediction will also be reduced
greatly.
This advancement aims to improve in the accuracy and reliability of current deterioration
modelling and to allow accurate future predictions to be calculated. And that is accurate
future predictions were made possible due to the optimised future validation values
resembling the real input values. The final result was a more reliable rate of deterioration.
This means a more accurate Bridge Management System could be developed. With future
prediction decay rates being more reliable, it is easier to calculate a bridge's expected life
more accurately
Stochastic models (Markov approach method)
The condition assessment of a structure reveals its state in relation to the structure’s
resistant to the loading capacity. That is overtime, the bridge will age to be different from
the original design and construction. Thus, development of bridge management system
needs to model the deterioration process to predict the trend of future degradation of
resistance. That is the rate of deterioration implies the prediction of the remaining lives as
well as future performance of the bridges or decay of the bridge performance over time.
Because deterioration is uncertain over time, it should ideally be represented as a stochastic
processes base on the Markov chain theory (JIANG et al 1989).

3. The Markov modelling is considered together with an optimal maintenance strategy is
developed by using the Markovian transition probability matrix for individual bridges can be
determined by considering bridge features and circumstances such as environment, width
and length of bridge, and traffic quantity.
Markov property is founded on two fundamental rules for transition probability: past
independency and homogeneous (Ng S. K., Moses F. 1998). That is past independency
suggest future states of the process depend only on the current state and not how the
current states had been reached. Homogeneous requires a rate of transition from one state
to another remain constant throughout the time. Consider a set of states, S = {s1, s2,…,sT}.
The process starts in one of these states and moves successively from one state to another.
The Markov-chain model assumes that a bridge can either remain in the current state or
deteriorate to the next lower state where the worst state space is considered an absorbing
state. By convention, all possible states and transitions have been included in the definition
of the processes, so there is always a next state and the process goes on forever If the chain
is currently in state si, then it moves to state sj at the next step with a probability denoted by
pij, and this probability does not depend upon which states the chain was in before the
current state. The probabilities pij are called transition probabilities. The process can remain
in the state it is in which is called holding time, and this occurs with probability pij.
The stochastic nature of deterioration process may be described as transition probability
matrix (P) is developed for the condition rating level 2 of individual bridge components and
valid with one transition matrix for the whole life span. The stochastic nature of the
deterioration process can be described in the format of:
[ ]
In order to develop a transition matrix, two approaches will be in consideration, the
frequency approach and the regression approach. By definition, frequency approach would
require at least two sets of inspection data pertaining to two different points in time. On the
other hand, regression approach only one set of bridge data is needed. The regression-
based optimisation method is the most-commonly used approach in estimating
transition matrices for different types of facilities, such as pavements and bridges (Bulusu
and Sinha, 1997). This method uses a non-linear optimization function to minimize the
sum of absolute differences between the regression curve that best fits the condition data
and the conditions predicted. In this investigation, calculate the probability transition
matrix from condition data is the percentage prediction method. This approach is can be
obtained directly from the condition data.

4. Level 2 bridge inspection model
The process of apply an inspection to assess and rate the condition of the structure to
identify the current maintenance needs. These inspections carry out a visual inspection of
bridge components including measurement of crack width (according with Bridge inspection
manual part 3.8.5) and an assessment of condition using a standard condition rating system
to represent the stage of deterioration. As such fundamental requirement to produce
consistent result shall be taken with state description as table below and classification of the
degree of deterioration with regard to environment affecting.
Condition
State
Subjective
Rating
Description
1 GOOD
(as new)
Free of defects with little or no deterioration evident
2 FAIR Free of defects affecting structural performance, integrity and
durability. Deterioration of a minor nature in the protective
coating and/or parent material is evident.
3 POOR
(monitoring
require)
Defects affecting the durability/serviceability which may require
monitoring and/or remedial action or inspection by a structural
engineer. Component or element shows marked and advancing
deterioration including loss of protective coating and minor loss of
section from the parent material is evident. Intervention is
normally required.
4 VERY POOR
(Remedial
action
require)
Defects affecting the performance and structural integrity which
require immediate intervention including an inspection by a
structural engineer, if principal components are affected.
Component or element shows advanced deterioration, loss of
section from the parent material, signs of overstressing or
evidence that it is acting differently to its intended design mode or
function.
5
(whole
structure
rating
only)
UNSAFE
(Immediate
Remedial
Action
Require)
This state is only intended to apply to the "whole structure"
rating. Structural integrity is severely compromised and the
structure must be taken out of service until a structural engineer
has inspected the structure and recommended the required
remedial action.
Exposure classification Location of component
Rating Environment
1 Relatively benign Interior of most structures and components above ground
on structures located more than 50km from the coast
2 Mildly Aggressive Components above ground in structures located between
1km and 50km from the coast or where components are in
contact with fresh water or soil.
3 Aggressive Components above ground within 1 km of the coast not
subjected to direct salt spray (ie. components in very damp
environments such as the wet tropics or rainforest areas),

5. and all components within 3m of permanent standing water.
4 Most Aggressive Components in tidal or splash zones or those subject to
direct salt spray or that are in contact with aggressive,
contaminated or salt rich soils*.
Data collection
The data was sourced from VicRoads of Victoria and includes the database of 3000 bridges
ready to be extracted detail of each critical component for the investigation. The
assessment will then be carried out for individual component and to display in the same plot
of graph. The evaluation of the behaviour in the graph concludes the pattern that allows a
prediction over unit of time. The data was formatted in a excel spread sheet including the
following information:
Structure id
Road Name
Feature Crossed
No of Spans Chainage
Start Reference Point
Distance Past
Road Number
MABC Classification
Road Classification
Region LGA
Latitude
Longitude
Year Constructed
Structure Form
Structure Type Bridge
Overall Length (m)
Overall Width (m)
Clear Width (m)
Traffic Width (m)
Date of Level 3 Inspection
Date of Level 2 Inspection
Date of Level 1 Inspection
Load Capacity (Tonnes)
High Mass Limit Constraint Y/N
Monitor Bridge YIN
Height Constraint Y/N
Other Agency / Responsibility
AADT
Sample of inspection data
Structure
ID
Date
Constructed
First
Inspection
Date:
Inspection
Condition Rating
Second
Inspection
Date:
Inspection
Condition Rating
Inspection
Interval
1 2 3 4 1 2 3 4
SN2086 30/6/1959 20/04/’04 100 0 0 0 06/06/'06 100 0 0 0 2.1
SN3936 30/6/1952 03/04/’03 82 5 11 2 06/06/’06 82 5 11 2 3.2
SN3232 01/6/1940 03/05/’02 100 0 0 0 24/01/’06 78 22 0 0 3.7
SN2800 30/6/1961 01/06/’02 100 0 0 0 04/04/’06 100 0 0 0 3.8
SN2104 30/6/1963 16/04/’04 69 10 21 0 24/1/’07 35 44 14 7 2.8

6. Reference list
1. BULUSU, S. & SINHA, K. C. 1997, “Comparison of Methodologies To Predict Bridge
Deterioration”, Transportation Research Record: Journal of the Transportation Research
Board, 1597, 9.
2. COLLINS, L. 1975, “An Introduction to Markov Chain Analysis”, Geo Abstracts.
3. Bridge Asset Management Structures Division Road Systems & Engineering, 2004. Bridge
Inspection Manual Second Edition.
4. Callow D, Lee J., Blumenstein M., Guan H., Loo Y. C. 2012, “Development of hybrid
optimisation method for Artificial Intelligence based bridge deterioration model-
Feasibility study”. ScienceDirect.
5. Cesare, M. A., Santamarina, C. J., Turkstra, and Vanmarke, E. H. 1992, “Modeling
Bridge Deterioration with Markov Chains”, Journal of Transportation Engineering,
ASCE, Vol. 118, No.6: 820-833
6. JIANG, Y. & SINHA, K. C. 1989, “Bridge Service Life Prediction Model Using the Markov
Chain”. Transp. Res. Rec., 24–30.
7. JIANG, Y. & SINHA, K. C. 1990. Final Report, Vol. 6: “Bridge Performance and
Optimisation”. West Lafayette, Indiana: Purdue University.
8. Huang, Y., H. 2010. “Artificial Neural Network Model of Bridge Deterioration”,
“Journal of Performance of Constructed Facilities”. ASCE
9. MADANAT, S., MISHALANI, R. & IBRAHIM, W. H. W. 1995, “Estimation of
Infrastructure Transition Probabilities from Condition Rating Data”, Journal of
Infrastructure Systems, vol. 1, pp. 120-125.
10. MADANAT, S. M., KARLAFTIS, M. G. & MCCARTHY, P. S. 1997, “Probabilistic
Infrastructure Deterioration Models with Panel Data”, Journal of Infrastructure
Systems, vol. 3, pp. 4-9.
11. MORCOUS, G. 2006 “Performance prediction of bridge deck systems using Markov
chains”, Journal of Performance of Constructed Facilities, vol. 20, pp. 146-155.
12. MADANAT, S., MISHALANI, R., and Wan Ibrahim, W. H. _1995_. “Estimation of
infrastructure transition probabilities from condition rating data.” J. Infrastruct. Syst.,
1_2_, pp. 120–125
13. MARK, A, C., CARLO, S., CARL, T., ERICK, H, V. 1992, “Modelling Bridge Deterioration
with Markvo Chains”. ASCE
14. Ng S-K., and Moses F. 1998, “Bridge deterioration using semi-Markov theory”. In
Harding, Shiraishi, Shinozuka &Wen (eds), Journal of Structural Safety and Reliability
pp. 113-120
15. SIANIPAR, P. R. M. & ADAMS, T. M. 1997. Fault-Tree Model of Bridge Element
Deterioration Due to Interaction. Journal of Infrastructure Systems, 3, 103-110.
16. Veshosky, D., Beidleman, C. R, Bueton, G.W and Demir, M., 1994,”Comparative
analysis of Bridge Superstructure Deterioration”, Journal of Structural Engineering,
ASCE, Vol. 120, No 7.